CONTRIBUTEDP A P E R
Growing Cells AtopMicroelectronic Chips:Interfacing Electrogenic CellsIn Vitro With CMOS-BasedMicroelectrode ArraysThis paper offers an overview of the fundamentals of bioelectronic measurements,
as well as the design, system integration, and application of CMOS-based microarrays
for recording from and for stimulating electrogenic cells from the brain and heart.
By Andreas Hierlemann, Member IEEE, Urs Frey,
Sadik Hafizovic, and Flavio Heer, Member IEEE
ABSTRACT | Complementary semiconductor–metal–oxide
(CMOS) technology is a very powerful technology that can be
more or less directly interfaced to electrogenic cells, like heart
or brain cells in vitro. To this end, the cells are cultured directly
atop the CMOS chips, which usually undergo dedicated post-
processing to obtain a reliable bidirectional interface via noble-
metal microelectrodes or high-k dielectrics. The big advantages
of using CMOS integrated circuits (ICs) include connectivity, the
possibility to address a large number of microelectrodes on a
tiny chip, and signal quality, the possibility to condition small
signals right at the spot of their generation. CMOS will be de-
monstrated to constitute an enabling technology that opens a
route to high-spatio–temporal-resolution and low-noise elec-
trophysiological recordings from a variety of biological pre-
parations, such as brain slices, or cultured cardiac and brain
cells. The recording technique is extracellular and noninvasive,
and the CMOS chips do not leak out any toxic compounds, so
that the cells remain viable for extended times. In turn, the
CMOS chips have been demonstrated to survive several months
of culturing while being fully immersed in saline solution and
being exposed to cellular metabolic products. The latter re-
quires dedicated passivation and packaging techniques as will
be shown. Fully integrated, monolithic microelectrode systems,
which feature large numbers of tightly spaced microelectrodes
and the associated circuitry units for bidirectional interaction
(stimulation and recording), will be in the focus of this review.
The respective dense microelectrode arrays (MEAs) with small
pixels enable subcellular-resolution investigation of regions of
interest in, e.g., neurobiological preparations, and, at the same
time, the large number of electrodes allows for studying the
activity of entire neuronal networks. Application areas include
neuroscience, as the devices enable fundamental neurophys-
iological insights at the cellular and circuit level, as well as
medical diagnostics and pharmacology.
KEYWORDS | Biomedical electronics; biological neural net-
works; bioelectric phenomena; complementary metal–oxide–
semiconductor (CMOS) integrated circuits; CMOS technology;
microelectrodes; mixed analog digital integrated circuits;
monolithic integrated circuits
Manuscript received December 2, 2009; accepted July 7, 2010. Date of publication
October 7, 2010; date of current version January 19, 2011. This work was supported
by ETH under the Internal Grant TH-00108, and by the Information Societies
Technology (IST) Future and Emerging Technologies program of the European
Union and the Swiss Bundesamt fur Bildung und Wissenschaft (BBW) under Contract
IST-2000-26463.
A. Hierlemann and F. Heer are with the Bio Engineering Laboratory, Department of
Biosystems Science and Engineering, ETH Zurich, CH-4058 Basel, Switzerland
(e-mail: [emailprotected]; [emailprotected]).
U. Frey was with the Bio Engineering Laboratory, Department of Biosystems Science
and Engineering, ETH Zurich, CH-4058 Basel, Switzerland. He is now with IBM Research
Zurich, 8803 Ruschlikon, Switzerland (e-mail: [emailprotected]).
S. Hafizovic was with the Bio Engineering Laboratory, Department of Biosystems
Science and Engineering, ETH Zurich, CH-4058 Basel, Switzerland. He is now with
Zurich Instruments, 8005 Zurich, Switzerland (e-mail: [emailprotected]).
Digital Object Identifier: 10.1109/JPROC.2010.2066532
252 Proceedings of the IEEE | Vol. 99, No. 2, February 2011 0018-9219/$26.00 �2010 IEEE
I . INTRODUCTION
A high degree of connectivity and the coordinated elec-trical activity of neural cells or networks are believed to bereasons why the brain is capable of highly sophisticatedinformation processing. Likewise, the effectiveness of ananimal’s heart also largely depends on such coordinatedcell activity. To advance our understanding of these com-plex biological systems, high-spatio–temporal-resolutiontechniques to monitor cell electrical activity and an ideallyseamless interaction between cells and recording devicesat both, single-cell and network levels are desired.
Methods to directly measure the electrical activity ofcultured electrogenic cells (e.g., neurons, heart cells,retina cells, or muscle cells) include two fundamentallydifferent techniques: 1) transmembrane measurements byaccessing the cell interior or cytosol with one of the elec-trodes, the so-called Bpatch clamp[ technique [1], [2], and2) extracellular recordings, e.g., by means of externalmicrotransducers [3]–[14]. Additionally, there are indirectmethods like optical measurements using voltage-sensitiveor fluorescent dyes [15]–[17]. The patch clamp techniqueyields very accurate information on the electrophysiolog-ical properties of entire cells, or, alternatively, on currentsflowing through single ion channels. However, it is aninvasive method and is limited in the cell viability time(usually hours) and in the overall number of cells that canbe simultaneously investigated.
For extracellular recordings, the cells are cultured orplaced directly atop the sensors or electrodes [Fig. 1(a)].When electrical activity occurs in a cell, ions flow acrossthe cell membrane within milliseconds. These moving ionsgenerate an electric field, which can directly influence theopen-gate region of a field-effect transistor (FET) [18]–[22], or which can be recorded by means of metalmicroelectrodes [23]–[31]. Extracellular recordings arenoninvasive (no puncturing of the cell membrane), which
enables long-term measurements. Multisite measurements
are possible by arranging many sensors in an array. For
stimulation, voltage transients can be applied by means of
electrodes or stimulation spots, which then evoke a depo-
larization of the nearby cell membrane and solicit
subsequent electrical cell activity. A stimulation spot is a
several-micrometer-diameter spot of either metal or di-
electric materials with high dielectric constant, such as
TiO2 ð" ¼ 34Þ and HfO2 ð" ¼ 15:4Þ, connected to an elec-
trical supply, through which stimulation signals are
applied [32]–[34].The most common biological preparations studied
using microelectrode arrays (MEAs) are divided into two
main categories: acute tissue preparations (e.g., slices),
which are recorded from immediately after they have been
removed from the animal [Fig. 1(b)]; cell cultures, which
can be further divided into dissociated cell cultures and
organotypic tissue cultures. Dissociated cultures are cells
whose relative in vivo positions are no longer preserved[see, e.g., Fig. 19(b)], whereas organotypic tissue cultures
are slices that are maintained in vitro over a period of time.
These variations encompass the spectrum from short-term,
functionally preserved tissue slices to long-term organo-
typic or cell culture preparations.
Several parameters must be taken into consideration
when using MEA chips as a tool for recording from cells.
Most importantly, the cell–electrode interface must beoptimized so that tight coupling between the cells and
electrodes is established, maximizing the signal-to-noise
ratio. The high-density microelectrode arrays (HD-MEAs)
described later (Section IV-E2) are capable of recording
from as deep in the tissue as 100 �m, but the signal-to-
noise ratio drops off rapidly with distance [35]. Acute
slices, for example, have a layer of dead cells that are
damaged during the cutting of the tissue, which introducesa degree of isolation between the active cells and the
Fig. 1. (a) Schematic of a cell attached to a sensor surface. A cell featuring ion channels sits on a planar microsensor (open-gate transistor or
electrode). Moving ions in the electrode vicinity generate an electric field or voltage that is recorded by the microsensor. (b) Micrograph of an
acute cerebellar brain slice (parasagittal cut) placed on a CMOS high-density electrode chip for measurements. The white regions of the slice
constitute myelinated areas, where no electrical activity can be recorded.
Hierlemann et al. : Growing Cells Atop Microelectronic Chips: Interfacing Electrogenic Cells In Vitro
Vol. 99, No. 2, February 2011 | Proceedings of the IEEE 253
electrodes. To promote the adhesion of cells or the tissuemass, the MEA chips are usually coated with cell-adhesion
substrates such as poly-D-lysine, laminin, collagen, and
fibronectin [36], [37]. Successful tactics for making
recordings from tissue sections utilize, for example,
cellulose nitrate or a coagulated plasma/thrombin clots
to stabilize the slices on the chip surface [38]. Further-
more, a perfusion system usually is necessary for the
delivery of carbogen gas (95% O2/5% CO2) and nutrients,which are required to sustain the physiologic processes of
the cells. Chip-through-holes have been realized on
circuitless chips to further improve nutrient access to the
bottom slice face on the chip and to thus enhance the
viability time [39]. This is very difficult if not impossible to
realize in high spatio–temporal resolution arrays due to
the densely packed circuitry and connection lines in the
chip as will be evidenced in Section IV.Substrate-integrated MEAs enable the study of neuro-
nal information processing [40]–[44] or the characteriza-
tion of cellular responses upon dosing biologically active
agents [25], [45]–[47]. Studying electrogenic cells in vitroprovides different experimental conditions in comparison
to investigating cells in vivo or in intact organisms: Most
animal and plant tissues comprise a variety of different cell
types, whereas a single specific cell type of interest orassemblies of selected different cell types can be grown in
culture. Moreover, the experimental parameters, such as
temperature or concentration of a chemical compound,
can be controlled much more rigorously in cell cultures as
compared to an organism. However, inferences from
in vitro-experiments to the in vivo behavior of cells have to
be experimentally verified. The investigation of acute or
organotypic slices, such as brain or other tissue slices[Fig. 1(b)] may help to bridge the gap between in vitro and
in vivo experiments to some extent [48], [49].
II . WHY USE IC OR CMOS TECHNOLOGY?
This review covers the application of complementary
semiconductor–metal–oxide (CMOS)-based, active micro-
arrays for recording from and stimulation of electrogeniccells in vitro. BActive[ array means that circuitry compo-
nents, such as amplifiers or filter units, are integrated with
the electrodes on the same substrate. The focus will be on
in vitro applications so that needle arrays carrying electrodes
and the respective in vivo applications will be omitted, as they
have been reviewed before in this journal [4], [50].
The three main advantages in using integrated circuit
(IC) or CMOS technology include most importantly1) connectivity; large numbers of transducers or electrodes
can be addressed by on-chip multiplexing architectures;
then 2) signal quality; the signal is conditioned right at the
electrode by means of dedicated circuitry units (filters,
amplifiers); and, less significantly, 3) ease of handling anduse; the devices and signals are robust (no Faraday cages,
digital signals, see below); many functions can be prog-
rammed or automated via user-friendly software anddigital interfaces that directly address digital registers
and logic or memory units on the chip side. Examples of
automation include self-identification of the chip, storage
of calibration or operation parameters, and the execution
of self-test functions. The presence of user-friendly fea-
tures is a crucial issue to encourage neuroscientists and
biologists to adopt such systems.
Traditional MEAs without multiplexers usually offer64 electrodes, with each electrode needing a connection to
external circuitry, which adds parasitic capacitance and
attenuates the small signals. CMOS-based MEAs, however,
comprise up to 16 384 electrodes with the necessary ad-
dressing circuitry on the same chip [49], [51]–[54]. The
high transducer density is an important asset, which
cannot be realized without on-chip multiplexers and ad-
dressing, simply since incorporating thousands of physicalor electrical connections on a millimeter-size chip is
impossible.
On-chip microelectronics provided by IC or CMOS
technology translate into improved system capability. For
example, signal conditioning can be performed on-chip,
ensuring that weak neural signals are faithfully recorded. It
is mandatory to not only amplify the signals but to also
limit the signal bandwidth to avoid noise aliasing in thesubsequent multiplexing. There are only very short
connection lines (a few micrometers) between electrode
and signal conditioning circuitry, and there are no long
lead lines running close to saline or the electrically active
culture under a thin layer of insulation. Therefore, any
interference with the target signals is largely obviated so
that Faraday cages and electrical shielding are not needed.
CMOS systems provide bidirectional communication viathe electrodes as they feature stimulation and recording
units. Active circuitry allows the stimulation of cells via
single or multiple electrodes in arbitrary configurations
from the set of densely packed electrodes, all while re-
cording from other electrodes without interruption. On-
chip analog-to-digital conversion (ADC) means that the
chip produces signals that may be easily manipulated and
transferred without compromising its information content.There is no influence of mechanical and electrical con-
nections (spring contacts) that cause problems for small
analog signals. Finally, the complete system can be mono-
lithically integrated on a single chip, which features an
area on the order of 20–40 mm2. The resulting measure-
ment setups are small, easily transportable and can be used
on a lab bench, or several of the devices can be placed in an
incubator with only a flat-band cable making connection tothe rest of the setup (see Section IV-E3).
As yet, the detailed structure of the extracellular-
potential landscape of a neuron and its dynamics are still
mostly unknown, since the signal-to-noise ratio and the
spatial resolution in the recordings of extracellular poten-
tials are, in most cases, low. This prevents a clear sepa-
ration of individual signal sources and an unambiguous
Hierlemann et al.: Growing Cells Atop Microelectronic Chips: Interfacing Electrogenic Cells In Vitro
254 Proceedings of the IEEE | Vol. 99, No. 2, February 2011
reconstruction of the neuronal networks derived from fieldpotentials or multiunit recordings [55]. HD-MEAs, real-
ized in standard CMOS technology, offer the potential to
perform recordings at single-cell or even subcellular reso-
lution [53]. The devices, however, have to meet two re-
quirements. They have to provide 1) high signal-to-noiseratio in the recordings of very small signals and, at the
same time, they have to enable 2) high spatial and temporalresolution measurements. The dilemma is that the two re-quirements are diametrically opposed in a technical reali-
zation. High spatial resolution entails the use of small
electrodes featuring higher thermal noise and that only
very few and small circuitry elements for addressing and
signal amplification can be realized per electrode. The
smaller the available space for electrodes and circuitry is,
however, the larger the noise is, since the noise of a tran-
sistor scales inversely with size, and since filters or am-plifiers requiring large capacitors cannot be realized. This
means that the signal quality is compromised by the cir-
cuitry generally needed to readout such high-density
arrays. Therefore, several devices to date feature a com-
parably high noise level that prevents the revelation of
electrophysiological details. There are, however, work-
arounds as will be shown in Section IV-E2.
A disadvantage of CMOS chips is that silicon is nottransparent to visible light in contrast to standard cell
culture substrates used in biology. Additionally, the chip or
its components may corrode upon operation and long-term
exposure. This holds particularly true as CMOS chips are
usually not made for being immersed in salt water over
extended times (weeks and months). Moreover, as soon as
stimulation features are implemented on-chip, severe
problems may arise from electrochemistry upon deliveringcurrent or voltage pulses and from the different redox
potentials of the variety of exposed materials, which may
generate local electrochemical elements and cause massive
corrosion (e.g., complete dissolution of CMOS metal alu-
minum). Therefore, a good packaging solution is needed,
on the one hand, to protect the chip against metabolic
products and chemicals of the cell culture, and, on the
other hand, to prevent the cells from being poisoned ordisturbed by toxic materials released by the chip, such as
the CMOS metal aluminum (also copper in modern pro-
cesses) that dissolves in saline solution. Viable solutions
will be presented in Section IV.
In the following, this review will first provide a short
introduction to the fundamentals of bioelectronics and
bioelectronic measurements (Section III), and then, an
overview on the design and application of CMOS-basedMEAs for interacting with electrogenic cells in vitro(Section IV). All systems feature the sensors or transducers
and the circuitry components on the same chip (monolithic
design). CMOS circuitry-only chips featuring recording
and/or stimulation units (see, e.g., [56] and [57]), which
are then connected to separate circuitless electrode array
chips (mostly on glass substrates), will not be covered here.
III . FUNDAMENTALS OF RECORDING OFELECTRICAL CELL ACTIVITY
A. Electrogenic CellsMany types of cells in the body have the ability to
undergo a transient electrical depolarization and repo-
larization that is either triggered by external mechanisms
(e.g., motor-nerve stimulation of skeletal muscle, or cell-
to-cell depolarization in the heart) or by intracellular,spontaneous mechanisms (e.g., cardiac pacemaker cells)
[58]. Cells that exhibit the ability to generate electrical
signals are called electrogenic cells, the most prominent of
which include brain cells or neurons and heart cells or
cardiomyocytes. Only a brief description of the biology of
electrogenic cells will be given in this section; for more
details the reader is directed to various books on this topic
[58]–[60].A huge number of neuronal cells form the nervous sys-
tem that regulates all aspects of body functions. The hu-
man brain contains about 1012 neurons (nerve cells) with
each neuron forming thousands of connections to other
neurons so that a large network of electrical connections
with massively parallel information processing character-
istics results. The output of a nervous system is the result
of its inputs and its circuit properties, that is, of the wiringor interconnections (synapses) between the single neu-
rons, and of the strength of these interconnections. The
synaptic connections between neurons can be reorganized,
which is known as synaptic plasticity, and is believed to be
a mechanism of learning in our brain. Different types of
neurons can be distinguished according to their physiology
and function in the body [58], [60].
Neurons exhibit four distinct regions with differentfunctions: the cell body, the dendrites, the axon, and the
axon terminals (Fig. 2). The cell body or soma with a dia-
meter of 10–30 �m contains the nucleus and is the pro-
duction site of most neuronal proteins. The majority of
neurons have a single axon, whose diameter varies from a
micrometer in the human brain to a millimeter in the giant
squid. Axons are specialized for the conduction of elec-
trical pulses, termed action potentials, away from the cellbody towards the axon terminals. Most of the neurons are
polarized: They receive synaptic input at the dendritic end
and provide output through synapses at the axonal end
(Fig. 2) [58].
Fig. 2. Schematic view of a neuron. Modified from [58].
Hierlemann et al. : Growing Cells Atop Microelectronic Chips: Interfacing Electrogenic Cells In Vitro
Vol. 99, No. 2, February 2011 | Proceedings of the IEEE 255
Neurons communicate with other neurons through
specialized contact zones, which are called synapses. Sy-napses generally transmit signals in only one direction,
from the axon terminal of the presynaptic cell to the den-
drites of the postsynaptic cell. Synapses can be either
electrical or chemical [58], [60]. Electrical synapses are
normally located at the gap junctions, which consist of
specialized proteins that form channels bridging the in-
teriors of two neurons and allow current flow from one
neuron to the other. In a chemical synapse, action po-tentials trigger the secretion of neurotransmitters in the
presynaptic neuron. The neurotransmitters then diffuse
across the synaptic cleft (20–25-nm width), and bind to a
receptor site on the postsynaptic membrane and cause
electrical signals there. The time required for this process,
the so-called synaptic delay, is typically 0.3–0.5 ms. The
synapse is an important feature for neural networks to
learn and to perform computation.Different ion concentrations inside and outside the
neuronal cell lead to an electric potential across the plasma
membrane (Fig. 3). In virtually all cases, the inside of the
cell membrane is negatively charged relative to the outside
[58], [60], i.e., a concentration gradient of ions between
the extracellular and the intracellular space (cytosol)
exists. Ion channels are embedded transmembrane proteins
that allow for the selective passage of certain ions. Ionchannels are termed gated if they can be opened or closed.
There are three types of gated ion channels: ligand-gated,
mechanically gated, and voltage-gated channels. The con-
centration of Kþ ions inside typical metazoan cells is about
ten times larger than that in the extracellular fluid, where-
as the concentrations of Naþ and Cl� ions are much higher
outside the cell than inside (Fig. 3). These concentration
gradients are maintained by membrane proteins that are
capable of pumping ions from one side of the membrane tothe other, often against their concentration gradients,
which consumes energy. There are additional types of ion
channels involved, but the membrane potential of most
electrically active cells is primarily affected by the opening
and closing of ion channels for Kþ, Naþ, Cl�, and
sometimes Ca2þ [58], [60].
The concentration gradient across the cell membrane
leads to an electrical potential, which is called restingpotential. The resting potential is mainly a consequence of
the membrane permeability to Kþ ions. The plasma mem-
brane contains resting Kþ ion channels, which are open in
the resting state and allow the passage of only Kþ ions. The
Kþ ions move across the membrane down their concen-
tration gradient, which leaves an excess negative charge on
the cytosolic face and deposits positive charge on the
exoplasmic face
Ex ¼RT
zxFln
cox
� �
dcixe: (1)
A variant of the Nernst equation (1) allows us to calcu-
late the maximum potential difference across a membrane
Ex (electromotive force), for any ion x at concentration cx.
Ex is given by the product of the thermal voltage RT=zxFand the concentration difference. R is the molar gas
constant, T is the temperature in Kelvin, zx is the number
of exchanged charges (1 for Kþ, Naþ, and Cl�), and F is theFaraday constant. The superscripts Bi[ and Bo[ denote the
ion concentration inside and outside the cell, respectively
[see (1)]
VM ¼RT
F� ln ½Ko�PK þ ½Nao�PNa þ ½Cli�PCl
½Ki�PK þ ½Nai�PNa þ ½Clo�PCl: (2)
The electric potential VM across a cell membrane can
be calculated by the Goldman–Hodgkin–Katz voltageequation (2), a more complex variant of the Nernst
equation, in which the concentrations of the ions are
weighted in proportion to the magnitudes of their trans-
membrane permeability constants P. P is a measure of the
ease with which an ion can cross a unit area of a membrane
driven by a one-molar difference in concentration; it is
proportional to the number of open ion channels and the
number of ions that each channel can conduct per second.Thus, PNa, PK, and PCl are measures of the Bleakiness[ of a
unit area of the membrane to these ions. The permeability
of the membrane to Naþ and Cl� ions is about one tenth
of that for Kþ, since there are about ten times more open
Kþ channels than open channels for Naþ or Cl�. In
theory, the resting membrane potential may be anywhere
between EK (�90 mV), the potassium equilibrium
Fig. 3. Intra- and extracellular ion concentrations for the species
primarily involved in the electrogenic properties of the neuronal
membrane, Naþ, Kþ, and Cl�. Redrawn according to [58]. Note that the
outside cell membrane is considered to be positively charged.
A� denotes all nonchloride anions (phosphates, etc.).
Hierlemann et al.: Growing Cells Atop Microelectronic Chips: Interfacing Electrogenic Cells In Vitro
256 Proceedings of the IEEE | Vol. 99, No. 2, February 2011
potential, and ENa (55 mV), the sodium equilibrium po-
tential. By inserting the typical ion concentrations into (2),
the potential of the resting membrane amounts to, e.g.,
�62 mV for the squid giant axon. Most neurons feature
resting potentials around �60 mV.
An action potential event is largely the result of
voltage-gated ion-channel conductances that cause a depo-
larization and repolarization of the neuronal plasma mem-brane within 2–3 ms [Fig. 4(a)]. The resting potential
across the membrane is, as already mentioned, maintained
by large transmembrane proteins that pump ions, against
their electrochemical gradient, in and out of the cell. The
action potential is triggered when the membrane is depo-
larized to a threshold value causing voltage-gated ion
channels to open. Naþ ions then flow down their elec-
trochemical gradient across the cell membrane (inwardcurrent, INa), further depolarizing the membrane. Depo-
larization causes Kþ ion channels to open, causing a cur-
rent flow in the opposite direction (outward current, IK)
that repolarizes the membrane [see Fig. 4(a)]. Finally, all
ion channels relax to their original state, and the mem-
brane returns, often after a brief hyperpolarization (mem-
brane potential more negative than resting potential), to
its resting potential. It has to be noted that the occurrenceof an action potential is an electrical event generated by a
change in the distribution of charge across the membrane
and not by a marked change in the overall intracellular or
extracellular concentration of Naþ or Kþ ions (ion con-
centration change of less than 1 per mill inside the cell).
Action potentials propagate unidirectional from the
axon hillock down the axon to the axon terminals (Fig. 2).
This is due to the fact that the voltage-gated Naþ channels
remain inactive for several milliseconds after opening, a
time span called the refractory period [58], [60]. The
inability of Naþ channels to reopen during the refractory
period also limits the number of action potentials per
second that a neuron can conduct.
Cardiac muscle cells, also denoted cardiac myocytes (or
cardiomyocytes), share similarities with skeletal muscle
cells with regard to their striated appearance and contrac-tion. A coordinated and synchronous activity of the cardio-
myocytes is necessary to effectively pump blood through
the vessels. Therefore, there is a direct connection be-
tween the cytoplasms of cardiac cells, called gap junctions,
which allows various molecules and ions to pass freely and
without attenuation. Accordingly, a depolarization wave
rapidly spreads through those direct connections without
characteristic temporal delay, which is observed forneuronal chemical synapses [58]. This direct cell-to-cell
connection means that one usually does not observe single-
cell action potentials, but so-called Bfield potentials[ upon
using extracellular electrodes to record from cell ensem-
bles. Field potentials are comparably large-amplitude
signals that result from simultaneous, collective electrical
activity of patches of electrogenic cells.
There are two general types of cardiac action poten-tials: 1) nonpacemaker action potentials, which are found
throughout the heart, and 2) spontaneous action potentials
of pacemaker cells. The latter are found in the sinoatrial
and atrioventricular nodes of the heart [6], [58]. Both
types of action potentials in the heart differ considerably
from those of neuronal cells in the duration of the action
potentials [neuronal cell about 1 ms; see Fig. 4(a); cardiac
action potential from 200 to 400 ms; see Fig. 4(b) and (c)]
Fig. 4. Schematic, time course, and current contributions of action potentials of different electrogenic cells as measured intracellularly.
Arrows pointing to the right represent ion currents flowing into the cell, arrows pointing to the left represent outward currents. (a) Neuron: Onset
of the action potential at 0.5 ms; the membrane potential becomes more positive (depolarization) through Naþ influx. After 1 ms return via
hyperpolarization state (potential more negative than resting potential) to resting potential through opening of Kþ channels: Kþ efflux.
Adapted from [58]. (b) Cardiomyocytes (ventricular myocyte): The numbers 0–4 mark the initial rapid depolarization phase (0), an initial
repolarization phase (1), a plateau phase (2), a rapid repolarization phase (3), and the membrane resting potential (4); for more information,
see text. (c) Action potential of a cardiac pacemaker cell: The numbers here mark the depolarization phase of the action potential (0),
a repolarization phase (3), and a spontaneous depolarization phase (4); for more information, see text.
Hierlemann et al. : Growing Cells Atop Microelectronic Chips: Interfacing Electrogenic Cells In Vitro
Vol. 99, No. 2, February 2011 | Proceedings of the IEEE 257
and the involved ion channels. In neuronal and nonpace-maker cardiac cells, the depolarization phase of the action
potential is caused by an opening of sodium channels,
whereas in cardiac pacemaker cells, calcium ions are in-
volved in the initial depolarization.
Nonpacemaker cardiac cells have a true resting mem-
brane potential [phase 4, Fig. 4(b)] that remains near the
equilibrium potential for Kþ ðEKÞ. When these cells are
rapidly depolarized to a threshold voltage of about�70 mVby another conducted action potential (phase 0), a fast
Naþ-current INa flows into the cell. At the same time, the
outward Kþ current IK is interrupted. These two changes
in ionic currents move the membrane potential farther
away from the resting potential and closer toward the
sodium potential. Phase 1 then represents an initial repo-
larization caused by the opening of a special type of Kþ
channel producing a transient outward Kþ current, IKðtoÞ.However, due to the large increase in the slow inward
Ca2þ current ICaðlÞ, the repolarization is delayed, and there
is a plateau phase in the action potential (phase 2). This
inward calcium movement is through long-lasting (l-type)
calcium channels that open up when the membrane
potential depolarizes to about �40 mV. Repolarization
(phase 3) occurs, when the outward Kþ current IK in-
creases again, and the Ca2þ current ICaðlÞ decreases.Therefore, the action potential in nonpacemaker cells is
primarily determined by changes in fast Naþ, slow Ca2þ,
and Kþ currents. During phases 0, 1, 2, and a part of
phase 3, the cell is refractory to the initiation of new action
potentials, i.e., the cell does not show any electrical
response.
Pacemaker cells are characterized as having no true
resting potential, but instead generate regular, spontane-ous action potentials [58]. There are, in fact, no fast Naþ
currents operating in those cells [Fig. 4(c)]. At phase 0, the
depolarization is primarily due to increased inward Ca2þ
currents ICaðlÞ. As the movement (or conductance) of Ca2þ
through the channels is not rapid, the rate of depolariza-
tion (slope of phase 0) is much slower than in other cardiac
cells. Repolarization occurs (phase 3) as the outward Kþ
current IK increases and the Ca2þ current ICaðlÞ decreases.Spontaneous depolarization (phase 4) is due to a fall in the
Kþ current IK and a small increase in the Ca2þ current
ICaðtÞ. A slow inward Naþ current also contributes to
phase 4, and is thought to be responsible for what is
termed the pacemaker or Bfunny current[ If . Once this
spontaneous depolarization reaches threshold (approxi-
mately �40 mV), a new action potential is triggered.
Although pacemaker activity is spontaneously generated,the rate of this activity can be modified significantly by
external factors such as autonomic nerves, hormones,
ions, and ischemia/hypoxia.
The rhythm of the heart is normally generated and
regulated by pacemaker cells within the sinoatrial (SA)
node, which is located within the wall of the right atrium.
SA nodal pacemaker activity normally governs the rhythm
of the atria and ventricles [6]. When this rhythm becomesirregular, too fast (tachycardia) or too slow (bradycardia),
or the frequency of the atrial and ventricular beats are
different, then this is called an arrhythmia [6]. Arrhyth-
mias can develop from either altered impulse formation or
altered impulse conduction. The former concerns changes
in rhythm that are caused by changes in the automaticity of
the pacemaker cells or by abnormal generation of action
potentials at sites other than the SA node. Altered impulseconduction is usually associated with a complete or partial
block of electrical conduction within the heart and
commonly results in reentry, which can lead to tachyar-
rhythmias. Many origins of arrhythmia, e.g., the so-called
QT-prolongation (Q and T denote specific features in the
waveform of an electrocardiogram), a delayed repolariza-
tion following the depolarization or excitation of the heart
[47], can be also observed in cell cultures, which opens alarge field of applications for MEAs in the pharmaceutical
industry (see also Section IV-E4) [25], [61].
B. Electrical Interface and Signal RecordingThe electrical interface includes the cell membrane in
physiological solution on the one hand and the transducer
on the chip on the other hand. The electrical properties of
the cell membrane are usually described by the gate model
of Hodgkin and Huxley [62]. The gate model has been
developed studying the squid axon and proposes that ion
currents are flowing through transmembrane channelsdown their concentration gradients. The channels feature
voltage-sensitive gating or gating particles. Fig. 5 shows the
equivalent circuit of the neuronal membrane according to
the gate model of Hodgkin and Huxley. The equivalent
circuit consists of parallel branches for the different ionic
contributions and the membrane capacitance. The tunable
resistors represent voltage-gated ion channels, and gx and
Ix denote the specific voltage- and time-dependent con-ductance for ion x and the current carried by ion x (e.g.,
Naþ, Kþ, and Cl�). The voltage sources arise from ionic
concentration gradients across the cell membrane, and the
Nernst potential Ex represents the equilibrium potential of
ion x according to (1). The main ionic currents include the
potassium current IK, the sodium current INa, and some
smaller currents summarized in a leakage current IL:. The
resulting total membrane current is then described by
IM ¼ CMdVM
dtþX
x
Ix
¼ CMdVM
dtþX
x
gxðVM; tÞðVM � ExÞ (3)
where IM is the total membrane current, CM is the mem-
brane capacitance, and VM is the membrane potential. For
Hierlemann et al.: Growing Cells Atop Microelectronic Chips: Interfacing Electrogenic Cells In Vitro
258 Proceedings of the IEEE | Vol. 99, No. 2, February 2011
more information on this model, the interested reader isdirected to [59] or, for cardiac cells, to [6].
As an example, neuronal electrical signals or action
potentials are shown for intracellular patch clamp mea-
surements (compare Fig. 4) and extracellular measure-
ments on the right-hand side of Fig. 5. The signals have
been recorded from the cell soma. The extracellularly
recorded signals feature an amplitude that is three orders
of magnitude lower than that of intracellular signals and abipolar shape with first a negative and then a positive peak
with respect to the baseline. The bipolar signal is similar to
that of lose patch clamp and can be explained with the
position of the electrode outside the cell: First, fast Naþ
current flows away from the electrode into the cell (nega-
tive peak), and, later, a somewhat slower current of Kþ
ions flows out of the cell towards the electrode (positive
peak). The intracellularly measured signals (Figs. 4 and 5)show almost opposite characteristics: A positive peak upon
the fast inward Naþ current and a small swing through
negative territory upon the outward Kþ current. In gene-
ral, there is a large variety of signal shapes and amplitudes
in extracellular recordings; examples will be shown in
Section IV-E4. These depend on the relative position of the
electrode with respect to the signal generating entity
(cluster of cells, single cell, soma, axon, dendrites), thenature of the cells under investigation, and the direction of
signal propagation. The analytical methods of electrical
cable theory are usually applied to investigate and explain
neuronal signal propagation [59]. For details on modeling
extracellular potentials and correlations between extracel-
lular and intracellular waveforms, the reader is referred to
[63] and [64].
The signal amplitude also depends on the cell size.Larger cells tend to yield higher signals, as the transmem-
brane currents that are needed to depolarize the cell arelarger in comparison to small cells. These currents depend
on the type and the number of contributing ion channels
and on the spatial distribution of the ion channels. Another
important issue is the distance between the transducer and
the neuron: the closer the cell is to the transducer, the
larger is the recorded signal. A third aspect includes the
resistance of the extracellular space to the different cur-
rents. A large resistance to spreading currents, or largesealing resistance, will entail high signals. The sealing
resistance Rseal in Fig. 5 can assume values between 1 and
10 M� [65]–[67]. The resistance to currents from the cell
to the electrode, i.e., the resistance of the gap between cell
and transducer Rgap, should be as low as possible and is a
function of the contact area, of the width of the gap, and of
the electrolyte resistance. In most dissociated cell cultures,
Rgap is much smaller than Rseal and can be neglected.Also important is the sensor/transducer size. Ideally,
the transducers should be as small as possible, since large
transducers will measure a potential averaged over a large
area, which reduces the peak signal amplitudes. On the
other hand, the use of smaller transducers may entail
higher noise levels in the signals, which may counterbal-
ance or even outbalance the small-size advantage. In dis-
sociated cultures, the cells are usually very close to theelectrodes, yielding large signals. Moreover, when large
neurons, such as snail neurons [68], are used, the signals
can reach up to tens of mV; in the case of cultures of
smaller mammalian neurons, they still can reach several
mV [69]. High-density cultures provide larger signals in
comparison to low-density cultures, as the sealing resis-
tance in the extracellular space is larger due to many
tightly spaced cells (ions have to move in the space belowthe cells along the transducer surface and cannot escape
Fig. 5. Electrical equivalent circuit of a cell membrane adjacent to a sensor, in this case, a metal electrode. The equivalent model of the
membrane relies on the Hodgkin–Huxley model of the squid axon [62]. The electrode is represented by a capacitor and resistor in parallel.
For details, see text. On the right-hand side, an intracellular neuronal signal recorded by using the patch clamp technique and an extracellular
signal recorded by means of a metal electrode are displayed for comparison, along with the involved ion currents. The Naþ-current flows into
the cell, and the Kþ-current flows outward. Note the three-orders-of-magnitude difference in signal amplitudes.
Hierlemann et al. : Growing Cells Atop Microelectronic Chips: Interfacing Electrogenic Cells In Vitro
Vol. 99, No. 2, February 2011 | Proceedings of the IEEE 259
into the open solution). Slices generally produce consid-
erably smaller signals, as a layer of cell debris is between
the intact cells and the electrodes.
The extracellular transducer or sensor, which records
the electrical signals, i.e., the potential changes upon ion
movement, is, in most cases, a metal electrode or an open-
gate field-effect transistor (OGFET); see Fig. 6 [70]. Thecell preparation is usually placed or cultured directly atop
the transducer element (Figs. 5 and 6), i.e., the transducer
is immersed in the physiological culture solution for days,
weeks, or months. The first extracellular recordings using
arrays of microelectrodes or OGFETs were performed in
the 1970s [20], [21], [30], [71].
Metal electrodes and OGFETs can be arranged in 2-D
arrays for in vitro applications. The same metal electrodescan be used for recording and for stimulation, since the
metal electrode/electrolyte impedance can be very low
(metal electrodes are not suitable for the measurement of
dc potentials due to electrode polarization [72]). An open
FET used for recording is, in most cases, accompanied by a
capacitive-type stimulation spot that then triggers electri-
cal cell activity [32]–[34]. In the case of the OGFET, firing
electrogenic cells change the extracellular potential in thegap between cell and the exposed gate oxide of the tran-
sistor (Fig. 6), which directly modulates the source-drain
current [Fig. 6(a)]. In common metal–oxide silicon (MOS)
FETs, the low-frequency noise is caused by electrons tun-
neling between silicon and traps in the gate oxide. This
noise is reduced when the electron channel of the tran-
sistor is buried in the silicon substrate, a few nanometers
away from the interface.The interface capacitance of an open-gate transistor
(electrolyte–oxide–silicon capacitance: 1.48 �F/cm2 for
TiO2 or 0.22 �F/cm2 for SiO2 [34]) is small in comparison
to that of a metal electrode (electrolyte–metal capacitance:
20–50 �F/cm2 depending on metal; see, e.g., [73]), which
renders stimulation more difficult due to the low charge
transfer capability. It also has to be noted that many of the
traditional OGFET devices need a strong biasing (positivebias with respect to the ionic solution) to avoid the pene-
tration of sodium and other ions in the open gate region,
which will diffuse into the device and alter the transistor
characteristics [74]–[78]. This problem has been alleviated
through the use of passivation layers and so-calledBextended-gate[ architectures as will be described below.
Metal electrode arrays have been used by several labo-
ratories to perform extracellular recordings [3], [9], [46],
[79]–[84], and electrode-based systems are commercially
available.1 These commercial systems mostly include passive
metal electrode arrays on glass or silicon substrates with
external signal conditioning and recording components.
The metal electrode/electrolyte interface is character-ized by a capacitive process resulting from redistribution of
charged and polar particles in the electrode vicinity with
no charge transfer between the solution and the electrode,
and a component resulting from the electron exchange
between the electrode and a redox species in the solution,
termed Faradaic process (Fig. 7). In a non-Faradaic scena-
rio, the electrode is typically highly polarized and behaves
almost as an ideal capacitor. For any electrode/electrolyteinterface, however, there is a range of potentials, in which
charged particles such as electrons are able to pass across
the interface. Charge transfer typically involves oxidation
or reduction reactions.
Platinum electrodes in physiological saline are gener-
ally modeled using both Faradaic and non-Faradaic pro-
cesses. The immersion of a metal in an electrolyte results
in the formation of space charge layers near the electrodesurface, the so-called Stern layer [85], which includes
layers of water dipoles or opposite-charge ions and their
hydration shells. These layers entail capacitive contribu-
tions to the overall electrode/electrolyte interface char-
acteristics. The distribution of the charges within a certain
distance from the electrode is taken into account using the
Gouy–Chapman model [85]. Fig. 7 shows the resulting,
commonly used equivalent-circuit model of the electrode/electrolyte interface. The model consists of a constant-
phase-angle impedance ZCPA, a charge-transfer resistance
RCT, the Warburg impedance ZW, and a serial resistance
Fig. 6. (a) Open-gate field-effect transistor (OGFET). (b) Extended-gate
field-effect transistor (EGFET) in CMOS implementation. Reprinted
from [70] and adapted from [51] with permission.
1Multi Channel Systems GmbH, Reutlingen, Germany (www.multi-channelsystems.com); Alpha MED Scientific Inc., Osaka, Japan(www.med64.com); Plexon Inc., Dallas, TX (www.plexon.com); AlphaOmega Co., Alpharetta, GA (www.alphaomega-eng.com); Bionas GmbH,Rostock, Germany (www.bionas.de).
Fig. 7. (a) Commonly used equivalent circuit model of the electrode/
electrolyte interface. (b) Simplified model, which is a suitable
approximation for extracellular recordings.
Hierlemann et al.: Growing Cells Atop Microelectronic Chips: Interfacing Electrogenic Cells In Vitro
260 Proceedings of the IEEE | Vol. 99, No. 2, February 2011
RS [86]–[88]. The empirical non-Faradaic constant phase-angle-impedance is given by ZCPA ¼ Kði!Þ�� where K is a
measure of the magnitude and � is a measure of the de-
viation from purely capacitive behavior ð� ¼ 1Þ. Typically,
� is between 0.7 and 0.9 [73], [86]–[89]. RCT is the
Faradaic charge transfer resistance. For platinum, the
charge-transfer resistance is in the range of several hun-
dred megaohms [73]. The Warburg impedance ZW takes
into account that ions, produced on the surface of theelectrode, need to diffuse away. This impedance is negli-
gible for large charge-transfer resistances, which produce
only small Faradaic currents. RS is the electrical resistance
between measuring electrode and counter electrode. This
resistance is on the order of kiloohms and can be neg-
lected. A simplified model version is given in Fig. 7(b),
which is also used as transducer or sensor equivalent cir-
cuit in Fig. 5. It includes a capacitor CE, in parallel to aresistor RCT. The time constant of this structure is given by
RCT � CE (on the order of 1 Hz or lower) and is below the
signal band of typical cellular signals (kilohertz range), so
that the electrode/electrolyte interface can be approxi-
mated by a capacitor.
The impedance of the microelectrode is an important
parameter for extracellular recording, since it determines
the noise of the electrode and the signal attenuation.Commonly used electrode materials include platinum
black [89]–[91], titanium nitride [26], [92], indium tin
oxide (ITO) [45], and iridium oxide [93]. Platinum black
and titanium nitride (regular columnar structure) form
high-surface area coatings with low electrode impedance,
which is beneficial with regard to noise performance.
Titanium nitride and iridium oxide feature comparably
high specific charge transfer capability, so that thesematerials are frequently used for stimulation electrodes
[92]. ITO is transparent so that the cells can be observed
from the bottom with standard microscopy tools for
biology. More recently, carbon nanotubes have also been
used as electrode material [94]–[97].
In addition to planar electrodes, 3-D designs have been
developed, such as pyramid-shape electrodes to achieve
better quality recording from slices as they may penetratethrough the outermost layer of dead cells [26], [27], or,
more recently, nail-shape electrodes, which are intended
to be internalized and overgrown by neurons or neurites
(see Fig. 8) [98], [99]. The nail-shape electrodes have been
applied to cell lines, but not yet to wild-type cells. It will be
very interesting to see which information can be extracted
from an array of nail-type electrodes, as the recording
scenario resembles that of multilocation patch clamp ex-periments [98] rather than that of planar multielectrode
arrays, where many neurons contribute to the signals
recorded from the respective electrodes.
Finally, it has to be noted that there is not a large
difference between the two transducer types, OGFET and
metal electrode, and the respective front-end readout
schemes in CMOS implementations. The transistor gates
are located at the bottom of the CMOS layer stack, assource and drain are diffused into the monocrystalline
silicon before the deposition of the dielectrics and metal
layers. To enable access to those transistor gates in order to
realize OGFETs, one would have to etch away all the layers
above the gate, which is occasionally done for chemical
sensors [100]–[102]. However, it is much easier to imple-
ment the so-called Bextended-gate[ FETs (EGFETs). In
EGFETs metal wiring through CMOS vias and the metallayers is used to electrically connect a metal electrode at
the surface of the CMOS layer stack down to the transistor
gate [Fig. 6(b)] [51]. To ensure a capacitive interface and
an Bopen-gate-like[ electrochemical situation, a dielectric
with high dielectric coefficient (TiO2, HfO2, ZrO2 [34]) is
afterwards deposited on the metal electrode. This dielec-
tric material is then in contact with the electrolyte solution
[Fig. 6(b)] [51], [68], [103]. For metal electrodes, inreturn, usually a filter capacitor (high-pass filter, compa-
rably large capacitance) is implemented in between the
metal electrode, which directly contacts the electrolyte
solution, and the first amplifying transistor (see Fig. 18 in
Section IV-E1) [104], [105]. So for both transducers, a
largely capacitive recording situation can be assumed, be it
through a layer of dielectric material at the interface
(EGFET), or through a CMOS filter capacitor (metalelectrode).
C. Extracellular StimulationStimulation capability is a crucial feature, especially in
the field of neuroscience, to expose the neuronal culture to
certain stimulus patterns, to repeatedly apply identical
patterns to achieve a Btraining[ of the neuronal colonies,
or to do closed-loop experiments [106]. To advance theunderstanding of network dynamics, a large number of
bidirectional electrical interfaces to the neuronal network,
i.e., stimulation and recording sites are needed. An ideal
system for stimulation includes the following features:
• possibility to stimulate any subset of electrodes;
• flexibility in the stimulation waveform;
• current and voltage stimulation capability;
Fig. 8. (a) Schematic representation of a neuron engulfing a
micronail-structured electrode. The tight junction between the cell
membrane and the sensor surface increases the sealing resistance
between cells and electrodes as compared to planar electrodes.
(b) Electron micrograph depicting a cross section through a PC12 cell
(neuronal cell line) engulfing three functionalized gold micronails.
The gold micronail in the center of the image was cut only through the
head, and the other two were cut along the micronail’s head and stalk.
Reprinted with permission from [98].
Hierlemann et al. : Growing Cells Atop Microelectronic Chips: Interfacing Electrogenic Cells In Vitro
Vol. 99, No. 2, February 2011 | Proceedings of the IEEE 261
• capability to do a fast switching from stimulation torecording even on the same electrode.
The so-called real-time all-channel stimulator (RACS)
by Wagenaar and Potter [107] based on discrete off-chip
components meets these requirements. The stimulation of
neuronal cells can be performed by means of current or
voltage pulses. Intracellular stimulation during patch
clamp measurements is usually done in the current
mode, since the stimulation of neurons in their naturalenvironment also occurs through current inputs to their
dendritic tree. For extracellular stimulation, both voltage
stimulation [13], [108], [109] and current stimulation [13],
[105], [110]–[112] methods are widely used.
For current stimulation, a voltage-limiting element is
required in order to prevent electrolysis, which may
change the local pH, produce undesired electrochemistry,
or damage the electrode. The applied current and, conse-quently, the number of charges released into the culture
can be controlled. However, beyond the electrode, the
exact current paths and the current density that arrives at a
targeted cell are not known. It has been found that in most
cases negative current pulses are very effective to excite
neurons to fire action potentials [107], [110]. In case of
voltage stimulation, the applied voltage at the electrode
during stimulation can be controlled so that no electrolysiscan occur as long as the voltage is small enough, but the
current flowing into the preparation is unknown. Biphasic
voltage-controlled pulses with the first part of the pulse
having a positive voltage sign constitute most effective
stimuli, since the sharp downward voltage transient be-
tween the positive phase and the negative phase corre-
sponds to a strong negative current pulse.
The pulse amplitude and its duration determine thestimulus efficacy [107], [111]. The number of cells that are
directly stimulated through a defined, voltage-controlled
pulse grows with the amplitude of the respective pulse
[107], [113]. The pulse must be strong and long enough to
allow the cell membrane and all parasitic capacitances in
the system to charge. Normally, the pulse amplitude is in
the range of hundred to several hundred microvolts, al-
though pulse trains of weak capacitive stimuli can also beused as has been published recently [32], [33]. Another
important parameter is the electrode impedance, since it
determines the current for a given voltage transient and
the voltage on the electrode for a given current transient,
respectively.
Electrical stimulation pulses are on the order of 1 V,
whereas the cell signals are typically on the order of
100 �V. The large disparity between the voltage ranges forstimulation and recording leads to artifacts in the recorded
signals, which can easily interfere with or completely mask
action potentials, even on electrodes that are rather distant
from the stimulation site. This stimulation artifact can be
removed by using software approaches [114], [115], which
are computationally expensive and often require real-time
processing. Moreover, hardware approaches, based on
active suppression circuits, have been successfully devel-oped (see also Section IV-E4b) [81], [104].
In a practical application, effective stimuli need to be
found that do not damage electrodes or cells and minimize
stimulation artifacts that interfere with the recording of
responses. For long-term experiments, it is crucial to pre-
vent damage to the electrodes and damage to the cells. The
latter may result from high charge injection or high charge
densities [116]. There is a couple of choices that have to bemade, which include 1) the electrode material; 2) the
electrode configuration, i.e., the selection of a specific
electrode or a set of electrodes to stimulate a target neuron
or neuronal colony; 3) the decision to use voltage or cur-
rent control; and 4) the selection of a stimulus waveform
and pulse duration (monophasic, biphasic, multiphasic or
asymmetric, etc.). For more details, see [111]–[119]. Any
stimulation will, however, affect the electrochemical equi-librium or, at least, the ionic distribution in the vicinity of
the stimulating transducer.
IV. INTEGRATED CMOS-BASEDSYSTEMS FOR IN VITRO APPLICATIONS
OGFETs in semiconductor technology have been devel-
oped in the group of P. Fromherz, Martinsried, Germany[3], [18], [120], and by Offenhausser et al. [121], [122]. A
CMOS-based approach making use of palladium (Pd) elec-
trodes has been reported on by Baumann et al. [123]. None
of these initial semiconductor-based devices exhibit any
on-chip circuitry or electronic components other than the
transducers themselves (FETs, electrodes).
As discussed in the introduction, the use of on-chip
microelectronics is imperative for the realization of largeror high-density multitransducer or multielectrode arrays
with stimulation and recording capabilities. High-density
multielectrode arrays, designed according to the pixel
concept (repetition of circuitry unit with each electrode),
which feature integrated multiplexers and in-pixel ampli-
fiers, have been presented by Berdondini et al. [91],
[124], [125]. Using the chip, one can simultaneously
record from 4096 electrodes featuring a 42-�m pitch. AnOGFET- or EGFET-based multitransducer design featur-
ing 16 384 transducers at 7.8-�m pitch was developed by
Eversmann et al. [49], [51]. This system features only a
single EGFET and two switching transistors per recording
site. Both high-density approaches exhibit limited area for
the in-pixel circuitry, and stimulation capability has not yet
been included. A variant of the OGFET-based chip with
stimulation features is under development. A CMOS chipwith on-chip stimulation capability, realized as an integ-
rated stimulation electrode located outside but close to the
recording-electrode array, has been presented by the group
of Kovacs at Stanford University [126]. Finally, monolithic
CMOS MEAs featuring 128 bidirectional electrodes in a
low-density design (pixel concept), or 11 011 bidirec-
tional electrodes in a high-density design (switch-matrix
Hierlemann et al.: Growing Cells Atop Microelectronic Chips: Interfacing Electrogenic Cells In Vitro
262 Proceedings of the IEEE | Vol. 99, No. 2, February 2011
approach as discussed in Section IV-E2) with full on-chipsignal conditioning and ADC have been presented by
Heer et al. [105], [127], [128] and Frey et al. [53], [129]. All
these systems will be detailed in the subsequent sections.
In general, high-density designs offer two major
benefits. 1) The number of successful cell-sensor contacts
is not influenced by cellular dynamics. Cells can be placed
randomly on the array and may move during a dynamic
network formation without escaping from the sensitivearea or electrodes. 2) Dense arrays with small pixels enable
subcellular-resolution investigation of regions of interest
in neurobiological preparations, and, at the same time, the
large number of electrodes allows for studying the activity
of entire neuronal networks.
A. 128 � 128 CMOS FET Array for ExtracellularRecording of Neural Activity
A CMOS-based array featuring 16 384 field-effect-
based sensors at a pitch of 7.8 �m has been developed by
the Max Planck Institute for Biochemistry, Martinsried,Germany, in collaboration with Infineon Technologies
Corporate Research, Munich, Germany [51], [68]. The
system setup and chip architecture are depicted in
Fig. 9(c).
The monolithically integrated sensor array includes
128 � 128 sensors or pixels. The column decoder period-
ically selects one of the 128 columns of the pixel array.
Moreover, control signals are provided by this decoder,which determine whether the pixels within the selected
column are operated in the readout mode or in a calibra-
tion mode [Fig. 9(a)]. Calibration of all pixels needs to be
performed before the pixels are operated in the readout
mode, and this calibration has to be periodically repeated
after a certain number of readout frames (typically 100).
This calibration procedure is needed as a consequence of
the mismatch of the transistor characteristics and of leak-age occurring at the gate of the transistors.
The full frame readout rate is 2 kf/s. Each of the
128 rows is connected to a separate readout amplifier
[Fig. 9(c)]. The outputs of these amplifiers are connected to
the chip output drivers or to dummy loads via sixteen 8-to-1multiplexers. The 16 output drivers provide an output
current to I/V converters arranged on an off-chip printed
circuit board. Finally, the buffered output voltages of the
I/V converters are processed by 16 ADCs with an effective
resolution higher than 8 b. Integrating larger analog signal
processing units in the pixel is not feasible if high spatial
resolution is required. Therefore, the pixel circuit shown in
Fig. 9(a) consists of only three transistors: One transistor isrequired for sensing the extracellular voltage, another one
for pixel selection. Due to the mismatch of the electrical
parameters of the relatively small sensor transistors, a cali-
bration transistor shorts the gate and drain of the sensor
transistors in the calibration mode of the pixel. The cali-
bration technique, which has to be continuously performed
during the measurements, is described in detail in [51]. The
disadvantage of this calibration procedure is that the re-sulting artifacts in the recording traces have to be sub-
tracted afterwards. The pixel requires 7.8 �m � 7.8 �m
area; the electrically sensitive area features a diameter of
4.5 �m so that an array of 128 � 128 pixels can be integ-
rated within an area of 1 mm� 1 mm as shown in Fig. 9(b).
In order to read out the 16 384 sensors of the array with a
reasonable number of lines, the column scanning is ex-
tended by a row multiplexing at a ratio of 1 : 8.Recording-site densities of conventional passive metal elec-
trode arrays (64 electrodes) amount to 25/mm2, whereas
this CMOS-based sensor array features 16 000/mm2.
The chips have been fabricated using a two-metal-layer,
n-well epi CMOS process, optimized for analog applica-
tions with a minimum gate length of 0.5 �m, a gate oxide
thickness of 15 nm, and a supply voltage of 5 V; for details,
see [51]. After this standard process (devices, two metallayers, nitride passivation, and tungsten vias) has been
completed, the specific postprocessing starts (compare also
Fig. 6 [51]). First, the surface is planarized by a chemical
mechanical polishing (CMP) step. An approximately
50-nm-thick Ti/Pt stack is then vapor-deposited and struc-
tured in a liftoff process. This metal stack is used to provide
the sensor electrodes and the adhesion layer for the contact
Fig. 9. High-density sensor array with 16 384 sensor pixels on 1 � 1 mm2. (a) Pixel circuit. (b) Blowup of the sensor array showing pixel
diameter and pitch. (c) Chip architecture. Reprinted from [51] with permission.
Hierlemann et al. : Growing Cells Atop Microelectronic Chips: Interfacing Electrogenic Cells In Vitro
Vol. 99, No. 2, February 2011 | Proceedings of the IEEE 263
pads. As sensor dielectric, a 40-nm-thick stack of 10-nm
TiO2, 5-nm ZrO2, 10-nm TiO2, 5-nm ZrO2, and 10-nm TiO2
layers is sputter-deposited at temperatures below 400 �C to
avoid damage to the CMOS metallization layers. The
measured effective relative dielectric constant of this stack
is approximately 45. The dielectric is removed by a reactive
ion etch on the bond pads, and a 400-nm-thick Au layer is
vapor deposited and structured using a liftoff process [51].The extracellular recording capabilities have been de-
monstrated, e.g., with neurons from a pond snail, lymnaea
stagnalis [51]. Fig. 10(b) shows the result of a neuron
contacted by an invasive microelectrode and stimulated by
the injection of a constant current of 0.1 nA for 500 ms [51].
The intracellular voltage is recorded with the same micro-
electrode as used for stimulation. Four action potentials
with amplitudes of 60 mV have been observed. The changeof the intracellular potential at the beginning and the end of
the current injection is a measurement artifact due to the
voltage drop, which occurs owing to the microelectrode
resistance. An area of 32� 40 pixels of the sensor array has
been selected for readout with a sample rate of 8 kHz. Data
of the extracellular potentials recorded from one pixel
underneath the neuron are shown in Fig. 10(b) [51]. Due to
the high-pass transfer function of the gap between cell andtransducer, the extracellular signal represents the deriva-
tive of the intracellular signal. The same chip has also been
used to record field potentials from cultured brain slices
[49]. The noise level has been found to range between 70
and 250 �VRMS [49], [51].
B. 64 � 64 High-Density Electrode Array forImaging Electrophysiological Activity
Another high-density array has been developed at the
University and the Centre Suisse d’Electronique et de
Microtechnique (CSEM), Neuchatel, Switzerland [54],
[91], [124], [125]. The CMOS design is based on a solid-
state active pixel sensor (APS) concept that was originally
developed for image sensors. The chip consists of an array
of 64 � 64 pixel elements on an overall active area of
2.5 mm� 2.5 mm. Each pixel occupies an area of 42 �m�42 �m and comprises a square-shape aluminum or gold
microelectrode of 21 �m � 21 �m as well as a pream-plifier. The center-to-center electrode pitch amounts to
42 �m. Based on image-video concepts, applied at the de-
vice and acquisition system level, full-frame data streams
from 4096 microelectrodes (�560 electrodes/mm2) are
acquired at 7.7 kHz. A subset of 64 electrodes can be read
out at up to 120 kHz. A dual-mode acquisition concept has
been implemented and enables to either acquire electro-
physiological activities from the full frame or to zoom onspecific region of interests (ROIs) at higher temporal
resolution [124], [125].
The in-pixel preamplifier has to fit into the small pixel
dimensions, so that only circuits with a low number of
transistors have been considered. A five-transistor opera-
tional transconductance amplifier [5-OTA in Fig. 11(b)],
optimized for low noise and small area, has been used for
the preamplifier in the first design [54], [130]. Three ad-ditional transistors (T1, T2, and T3) have been integrated
in addition to the 5-OTAs in each pixel as can be seen in
Fig. 11(c). The presence of these additional transistors
allows for operating the preamplifier in open- or closed-
loop mode. In the open-loop mode, the signal is directly
amplified by the open-loop gain, while in the closed-loop
mode, the gain is set to one, and the preamplifier acts as an
impedance transformer.The most recent architecture of the overall system in-
cludes 1) the 64 � 64 MEA, on-chip amplification and
Fig. 10. (a) Snail neuron on the sensor chip in culture. (b) Measured data from a snail neuron. The neuron is stimulated and monitored
for reference purposes with a micropipette. Stimulation current ISTIM, intracellular potential VINTRA, and extracellular potential VEXTRA,
as recorded with the sensor array. Reprinted from [51] with permission.
Hierlemann et al.: Growing Cells Atop Microelectronic Chips: Interfacing Electrogenic Cells In Vitro
264 Proceedings of the IEEE | Vol. 99, No. 2, February 2011
filtering circuits as well as a high-speed random addressing
logic; 2) a field-programmable gate array (FPGA)-based in-terface providing analogue-to-digital converters (12-b reso-
lution), addressing control, real-time data preprocessing, and
a serializer for sending the data to a frame grabber through a
high-speed interface; and 3) a host computer equipped with a
frame grabber and running a custom software application
managing the platform control, data acquisition, and data
analysis on a multicore processor [54], [125].
Electrophysiological measurements have been per-formed by culturing dissociated hippocampal neuronal
cultures, prepared from embryonic E18 rats, on the alu-
minum electrodes of the chip. In the most recent gene-
ration of the chips [54], there is no noble metal layer on
the electrodes, whereas, in previous versions, gold has
been used as electrode material [130]. The CMOS
passivation has been opened and the topmost CMOS
aluminum layer is directly exposed to the cell culture. It isremarkable that, according to the authors, the CMOS
aluminum withstands the culturing and the cell metabolic
activity, and, on the other hand, that the cells tolerate the
presence of the aluminum for extended time [54].
Measurements have been conducted between day 14 and
day 30 in vitro. The cells showed spontaneous electrical
activity featuring amplitudes (extracellular potentials)
between 100 and 400 �VPP. Fig. 12 shows the recordedspontaneous neuronal bursting activity recorded from the
whole array. The noise of the system is approximately
11–15 �VRMS. The spatial and temporal resolution of the
device enables visualization of local spike propagations aswell as whole network (bursting) activities [54], [125]. The
chips have been reported to be reusable after cleaning the
devices with isopropanol and rinsing in de-ionized water.
The high-density system is commercially available from the
company B3Brain,[ Landquart, Switzerland.2
C. Multiparameter Sensor ChipThe integration of several different sensor types on one
chip for monitoring not only electrical cell activity, but
also temperature, cell metabolism, and cell adhesion has
been developed by the University of Rostock in coopera-
tion with the semiconductor company Micronas GmbH,Freiburg, Germany3 [123], [131]. The aim is to monitor
electrical signals as well as metabolic parameters with one
sensor chip. The applications are in the field of basic
research and drug screening.
The developed system, the concept of which is illus-
trated in Fig. 13, provides online monitoring of cellular
reactions under well-controlled experimental conditions.
The system includes cell-potential field-effect transistors(CPFET, sensitive gate areas of 6� 1 �m2) and palladium
electrodes (10-�m diameter, palladium features low cell
toxicity according to [133]) to measure the electrical cell
2http://www.3brain.com/.3http://www.micronas.com.
Fig. 11. (a) Overview of the high-density electrophysiological platform. The system includes the CMOS-MEA chip, the interface board,
and a work station equipped with a frame grabber for capturing and storing the data. Reprinted from [54] with permission. (b) Schematic of
the in-pixel preamplifier. (c) Detailed schematic of the five-transistor operational transconductance amplifier (OTA) used in the preamplifier.
Reprinted from [130] with permission.
Hierlemann et al. : Growing Cells Atop Microelectronic Chips: Interfacing Electrogenic Cells In Vitro
Vol. 99, No. 2, February 2011 | Proceedings of the IEEE 265
activity as shown in Fig. 14(a) [123], a sensor to monitor
the temperature of the cell culture, and ion-sensitive field-
effect transistors (ISFETs) to monitor the pH in the cel-
lular microenvironment, a recording of which is shown in
Fig. 14(b) [132]. The ISFETs allow for monitoring local
acidification and respiration in in vitro cell networks. The
interdigitated electrodes are used to measure the cell ad-
hesion by means of impedance measurements [134]. Cell
health can be characterized by the degree of adhesion of
electrically active cells to transducers. According to the
authors, impedimetric measurements using interdigitated
electrode structures have been found to provide informa-
tion on the cell density and number, the cell adhesion, and
the cellular morphology, since an alternating current (ac)
Fig. 12. Acquired activity during a propagating burst event. Data from a rat hippocampal neuronal culture (31 DIV, 7.7-kHz sampling,
4096 electrodes). View of the overall active area as an image sequence. Each frame is computed at a different time and shows the burst
propagation over the network. Reprinted from [125] with permission.
Fig. 13. Cell monitoring system concept: thermoregulated cell culture chamber with fluid handling system and different microsensors
(ISFET: ion-selective FET; ENFET: enzyme FET; ISE: ion-selective electrode; CPFET: cell potential FET; TD: temperature diode;
CCD: charge-coupled device) Reprinted from [132] with permission.
Hierlemann et al.: Growing Cells Atop Microelectronic Chips: Interfacing Electrogenic Cells In Vitro
266 Proceedings of the IEEE | Vol. 99, No. 2, February 2011
between the electrodes is influenced by the presence andstructural properties of living cells growing on these elec-
trode structures. The system as shown in Fig. 13 is com-
mercially available through Bionas GmbH, Rostock,
Germany.4 Toxicology investigations with this system
have been described, e.g., in [135]. More information on
how different chemicals trigger cellular responses of pre-
vailingly electrogenic cells can be found in the literature
[45], [46], [90], [126], [132], [136]–[138].
D. Portable Cell-Based BiosensorField-portable and handheld cell monitoring systems
have been developed at the Department of ElectricalEngineering, Stanford University [14], [126]. To realize
field-portable detectors for chemical and biological agents,
considerable changes must be made relative to the equip-
ment used in typical laboratory-based situations. First, the
entire system must be robust, relatively compact, and able
to be self-powered for a reasonable operational period.
Second, the cells must be packaged in a way as to maintain
sterility and proper environmental conditions for the cul-ture, and yet allow the introduction of unknown agents.
Typically, the cells, MEAs, and potentially some fluidic
interfaces are combined into a user-replaceable cartridge.
The cells must also be maintained at constant pH, osmo-
larity, and temperature, typically via chemical buffering,
humidification, and closed-loop thermal control [12], [46],
[139]. The high sensitivities of cells to changes in these
parameters can lead to false Bsignals[ if they are not tightlycontrolled. In addition, the sample preparation, which is
typically done manually in the laboratory, must be auto-
mated so that it can be carried out under realistic envi-
ronmental conditions.
Since the first such work presented in [46], majorefforts toward realizing field-portable systems have been
undertaken [139]. As an example, a handheld recording
system (Fig. 15) with integrated cell cartridges incorporat-
ing closed-loop thermal control and front-end amplifica-
tion and multiplexing electronics was developed [12], [14].
A die microphotograph of the CMOS chip used in this
portable device with a floor plan overlay is shown in
Fig. 15(b). Each chamber has one large pseudoreferenceelectrode, two stimulation electrodes, four arrays of
16 microelectrodes, multiple temperature sensors, and
nine 16-channel multiplexers. Digital logic, which uses a
three-wire serial interface, controls the multiplexers. The
poly(dimethylsiloxane) gasket seal contacts the top surface
of the die except for the two elliptical chamber areas and
the bond pads. The temperature is regulated by an on-chip
temperature controller. The goal of the temperature regu-lation system is to maintain the substrate beneath the cells at
a defined temperature even when exposed to environmental
transients. An n-channel metal–oxide–semiconductor
(NMOS) transistor is used as the heating element [Fig. 16(a)].
The primary biological sensors of the cell cartridge are
multiple arrays of 16 gold-coated planar microelectrodes
of 10-�m diameter. Both chambers have four arrays of
16 electrodes, resulting in 128 total sensor channels on asingle cartridge. Each microelectrode has a connection to
two separate multiplexers. One connection leads directly
to a multiplexer (for platinization, impedance measure-
ments, and unbuffered action potential measurements),
while the other connection goes to a low-noise source
follower, the output of which is routed through the second
multiplexer [for buffered action potential measurements;
Fig. 16(a)]. This configuration allows for simultaneousrecordings from four buffered and four unbuffered micro-
electrodes per chamber. A large pseudoreference/counter4http://www.bionas.de.
Fig. 14. (a) Extracellular recordings from one of the chip electrodes (approximately 40 superimposed neuronal action potentials).
Reprinted from [123] with permission. (b) Extracellular acidification measurements in a neuronal network on a silicon chip as performed
with ISFETs in a flow-through system. The pump cycle was 5 min ‘‘pump on’’ and 10 min ‘‘pump off.’’ During the pump-off period, the pH of the
medium decreased significantly due to the acidification through the presence of the cells. In the pump-on period, fresh medium was pumped
through the chamber. Reprinted from [132] with permission.
Hierlemann et al. : Growing Cells Atop Microelectronic Chips: Interfacing Electrogenic Cells In Vitro
Vol. 99, No. 2, February 2011 | Proceedings of the IEEE 267
electrode in each chamber is used to hold the solution at a
nearly constant potential during the measurements. The
analog portion of the handheld device includes low-noise
amplifiers, the auxiliary channel amplifiers, and ADCs(temperature, system power, reference voltage, battery
voltage), the stimulation system, and a pump control sys-
tem. The low-noise amplifiers connect to the eight chan-
nels of buffered source-follower cartridge outputs. The
amplifier system was designed to provide a gain of 505
across a 16 Hz–3 kHz bandwidth with 1 �VRMS of input-
referred noise.Signals from HL-1 cells (cardiomyocyte cell line) were
recorded at eight days in vitro [Fig. 16(b)]. The typical
amplitude of the recorded signals was in the 400 �VPP
Fig. 16. (a) Circuit block diagrams of the CMOS cell cartridge. Temperature control system showing components and off-chip connections (top)
and block diagram of one microelectrode sensor channel showing the unbuffered and source–follower–buffered connections (bottom).
(b) Extracellular action potential waveforms recorded with the handheld system from HL-1 cells (cardiomyocyte cell line) growing for eight days
in chamber A ðVA2Þ and chamber B ðVB3Þ of a cell cartridge. These signals were averaged 16 times and are of typical shape and amplitude
(300–400 �VPP). Reprinted from [12] with permission.
Fig. 15. (a) Handheld cell-based biosensor system. Front view showing the cartridge, pump, display, and control buttons; back internal view
showing the electronic subsystems on the right. (b) Microphotograph of the 6� 9 mm2 die. The lightened elliptical shading was added to
highlight the cell chamber surface areas. Each chamber has four arrays (quadrants) of 16 microelectrodes each. The serial digital interface
controls the eighteen 16-channel multiplexers. Reprinted from [12] with permission.
Hierlemann et al.: Growing Cells Atop Microelectronic Chips: Interfacing Electrogenic Cells In Vitro
268 Proceedings of the IEEE | Vol. 99, No. 2, February 2011
range with the largest observed signal amplitudes being1.3 mVPP.
E. Fully Integrated Bidirectional Low-Density andHigh-Density Electrode Systems
Two types of planar monolithic microsystems in CMOS
technology that provide bidirectional communication (sti-
mulation and recording) to cultured electrogenic cells or
slice preparations are presented in this section [52], [53],
[104], [105], [128]. Both microsystem chips are directlyused as a substrate for cell culturing, they feature circuitry
units for stimulation and immediate cell signal treatment,
and provide on-chip signal transformation as well as digital
interfaces, so that a very fast, almost real-time interaction
(2-ms loop time from event recognition to a defined sti-
mulation) is possible at good signal quality. The low-
density system (Section IV-E1) has the circuitry units
repeated with each electrode (pixel approach), whereas forthe high-density system (Section IV-E2), a switch matrix
strategy has been pursued, and the electronics have been
shifted to chip space outside the array in order to achieve
better lateral resolution and, at the same time, high signal
quality. Sample recordings of spontaneous and stimulated
electrical activity with neuronal and cardiac cell cultures
will be shown in Section IV-E4.
1) Low-Density 128-Electrode System: A micrograph of the
monolithic low-density CMOS microsystem is shown in
Fig. 17 [105]. The 6.5 mm � 6.5 mm chip comprises
128 stimulation- and recording-capable electrodes in an
8 � 16 array and an integrated reference electrode. The
system is structured in a modular design (Fig. 17). Each
pixel of the microelectrode array incorporates the signal-transducing electrode, a fully differential bandpass filter
for immediate signal conditioning, a mode storage unit,
and a buffer for stimulation. The pitch of the pixel units is
250 �m.
Electrode pitch, size, and shape are very flexible, and
the electrode material can be selected from a large variety
of materials, since the electrodes are realized during the
post-CMOS processing, as described in [104] and [105]. Adigital control unit is also integrated on the chip and con-
trols the multiplexing, the electrode selection for stimula-
tion, and the reset of single electrodes. Also it contains the
successive-approximation registers of the ADCs and the
interface to the outside world. The overall setup includes,
besides the chip, an FPGA and a laptop; see Section IV-E3.
Implementing filters and buffers at each electrode
offers important advantages. 1) The signal is amplified andfiltered in close proximity of the electrodes, which makes
the design less sensitive to noise and interference picked
up along connection lines. 2) One stimulation buffer per
electrode ensures that the commanded stimulation wave-
form can be reliably transmitted to each selected electrode
and that it is independent of the overall number of
electrodes selected for stimulation. 3) The high-pass filter
removes offset and drift of the biochemical signals so thatthe signal can be amplified before it is multiplexed. 4) The
low-pass filter (LPF) limits the noise bandwidth and works
as an antialiasing filter for the multiplexing and for the
subsequent ADC.
The schematic of the in-pixel circuitry components,
which have been optimized for low noise and small area, is
shown in Fig. 18. The first readout stage is a high-pass filter
Fig. 17. Micrograph of the 128-electrode CMOS system chip (6.5 � 6.5 mm2) and closeup of one pixel. The different components include the
128-electrode array (8 � 16) featuring amplification and filtering circuitry units associated with each electrode; 16 ADCs; digital control and
interface; DAC for stimulation; temperature sensor. The closeup shows one pixel repeating unit featuring the two CMOS aluminum contacts,
one for readout, one for stimulation, and the final platinum electrode shifted sideways from those contacts (see also Section IV-E3).
Hierlemann et al. : Growing Cells Atop Microelectronic Chips: Interfacing Electrogenic Cells In Vitro
Vol. 99, No. 2, February 2011 | Proceedings of the IEEE 269
to eliminate drift and offset of the metal electrode. The gain
of the filter is 20 dB defined by the capacitance ratio C1=C2.
The corner frequency is given by the capacitor C2 and theresistor R. The value of C2 is a tradeoff between gain accuracy
and corner frequency. Given the spatial constraints imposed
by the 250-�m pitch, a 200-fF capacitance was chosen for C2.
To realize a corner frequency of 1 Hz, a large resistance R is
required. Such large resistance has been realized by using a
MOS resistor (see [140]) with a gate length of 100 �m and a
gate width of 1 �m. The absolute value of the resistor can be
externally controlled by changing the gate voltage. Thecorner frequency can be tuned from approximately 1 Hz to
about 1 kHz. A second design that makes use of a MOS diode
to realize such large resistances has been implemented as a
test structure (see [141] and [142]). For the MOS-diode-
based approach, no reference voltage is required, but the
filter stage is not tunable as it is with the MOS resistor. The
corner frequency of the diode-based design was measured to
be approximately 1 Hz.In the subsequent filter stage, a passive MOSFET-C
LPF limits the noise bandwidth and prevents aliasing. The
area (90 � 45 �m2) of the passive LPF was minimized by
using MOS resistors that are tunable by modification of the
gate voltage and by using MOS capacitors. The corner
frequency of the filter can be tuned from about 1 to
30 kHz. Following the LPF, an additional gain stage am-
plifies the signals and allows for fast multiplexing. Thisbuffer is a fully differential version of the amplifier pre-
sented in [143] and has a gain of 30 dB.
The in-pixel circuitry unit also includes stimulation
capabilities. Any arbitrary stimulation pattern (with a
maximum sampling rate of 60 kHz and 8-b resolution) canbe applied to any electrode. The readout circuitry at each
electrode can individually be reset to its operating point in
order to suppress artifacts evoked by the stimulation pulses
from the stimulated electrode itself or from neighboring
electrodes. The stimulation buffer (Fig. 18) comprises a
differential input stage ðM1;M10Þ with an active load ðM2Þand is connected in a unity gain configuration (OUT con-
nected to gate of M10). The class-AB stage is formed by thetransistors M3, M4, M5, and M6, where M3 and M4, both
of which are diode connected, bias the output transistors
M5 and M6 [89]. The power consumption is 45 �W for
very small inputs at a current Ibias of 2 �A and a supply
voltage of 5 V. However, when slewing, the amplifier can
deliver up to several milliamps to the electrode. The whole
buffer occupies an area of 42 �m � 32 �m. Transmission
gates are used as switches (Stim) to connect the buffer tothe electrode. For more details on the circuitry see [104]
and [105].
A total equivalent input noise of the pixel circuitry of
11.7 �VRMS (0.1 Hz to 100 kHz) has been measured. The
electrodes are continuously read out at a sampling rate of
20 kHz per electrode. A gain of 1000 or 3000 can be
selected. The overall power consumption of the chip is
120 mW at 5-V supply. Electrogenic cells are very sensitiveto temperature, so that even subtle temperature changes
may change the cell activity level and may even lead to cell
Fig. 18. Circuitry unit, which is repeated at each electrode. It comprises a stimulation buffer, a high-pass filter, an LPF, and a final amplification
stage. The low-frequency corners of the filters are realized by using MOS resistors. The cutoff frequency of the filters can be adjusted by
changing the gate voltages (Vb_HPF and Vb_LPF) of the MOS resistors. Reprinted with permission from [105].
Hierlemann et al.: Growing Cells Atop Microelectronic Chips: Interfacing Electrogenic Cells In Vitro
270 Proceedings of the IEEE | Vol. 99, No. 2, February 2011
death. Therefore, excessive power consumption and heat
dissipated by the electronics directly underneath the cellshas to be avoided or accounted for by regulating the tem-
perature of the culture liquid. An on-chip temperature
sensor has been implemented to monitor the chip
temperature. Operating the chip with liquid on the surface
leads to a temperature rise of less than 1 �C with respect to
ambient temperature upon operation, so that additional
cooling of the system is not required. Circuits operating at
low frequency (down to 1 Hz) might be sensitive to leakagecurrents, the effect of which has been reduced by the fully
differential design of the in-pixel readout circuitry.
Furthermore, electromagnetic coupling is also generally
reduced in a fully differential architecture.
2) High-Density 10 011-Electrode System: A workaround
for the dilemma of achieving high signal-to-noise ratios
and, at the same time, high lateral resolution in the re-cordings (see Section II), is presented here [52], [53],
[129]. Instead of simultaneously reading from all electro-
des, which requires the front-end amplifiers to be located
directly at each recording electrode, the workaround in-
cludes the realization of a matrix of switches and memory
cells (little area needed and negligible noise contribution)
underneath the electrodes to route a subset of these
electrodes to circuitry units (channels) placed outside thearray, where no area constraints apply [52], [53], [129].
This way, high-density electrodes are combined with high-
performance circuitry, which features noise levels as low
as 3 �VRMS. It is important to mention that the imple-
mentation enables an almost arbitrary selection of elec-
trodes (cohesive blocks, lines, single electrodes here and
there) to be connected to the 126 available readout chan-nels. The electrode selection, which can be changed within
approximately 2 ms, can be adapted to the biological-
sample structure, and one can do, e.g., a large-area activity
screening or mapping of the functional connectivity and
then proceed to a detailed study of a certain area or net-
work. The large redundancy in the data gained from such
detailed recordings in the area of a single cell can be
effectively used for spike sorting to enable a precise, tem-porally and spatially highly resolved reconstruction of, e.g.,
the extracellular potential landscape of a single cell, which
can be used to test and validate corresponding models.
The microsystem chip is 7.5 � 6.1 mm2 in size, and
features 11 011 metal electrodes as well as 126 bidirec-
tional circuitry channels, each of which is equipped with
recording and stimulation electronics [Fig. 19(a)]. The
electrodes feature a diameter of 7 �m and are placed at apitch of 18 �m (honeycomb pattern) in an area of 2.0 �1.75 mm2 yielding a density of 3150 electrodes/mm2
[Fig. 19(b)].
Flexibility in the electrode selection is attained by an
analog switch matrix, integrated underneath the electrode
array, which consists of 13-k memory cells and analog
switches to define the routing from the electrodes to the
amplifiers of the channel units [52], [53], [129]. In brief,the memory cells store the information whether the re-
spective switch is closed or not, i.e., whether an electrode
is connected via several switches to one of the readout/
stimulation channels. To obtain the settings of the
switches for a selected set of electrodes, the switch matrix
Fig. 19. (a) Micrograph of the MEA chip (7.5�6.1 mm2). The electrode array is surrounded by the first and second amplification and filtering stages
and the stimulation buffers. Below the array is the shift register used to program the array and on the right-hand side are the third amplifier
stages, ADCs, and the digital core. (b) Colored scanning electron micrograph of chicken dorsal root ganglion neurons (two days in vitro) cultured on
the chip illustrating size and density of the microelectrodes (blue) in comparison to the neurons (green). Reprinted from [53] and [129] with
permission.
Hierlemann et al. : Growing Cells Atop Microelectronic Chips: Interfacing Electrogenic Cells In Vitro
Vol. 99, No. 2, February 2011 | Proceedings of the IEEE 271
is represented as a graph, and a max-flow, min-cost opti-mization problem is solved using an integer-linear prog-
ram (ILP). Arcs from spots of interest to a set of electrodes,
on which the signal can be measured, are assigned a cost,
such as the Euclidean distance or a value that is inversely
proportional to the signal quality. Simulations for the task
of routing 126 randomly distributed spots to the 126 read-
out channels using the Euclidean distance as a cost mea-
sure were carried out to assess the ability to select spots ofinterest. The average distance from any spot of interest to
its closest electrode was 6.75 �m for the set used
ðn ¼ 2000Þ. The implemented routing scheme provides
an average distance to the connected electrode of 7.1 �m,
where 114.6 of the 126 spots can be read out via the closest
electrode. One hundred two electrodes in a 6 � 17 rec-
tangular configuration constitute the largest obtainable
coherent electrode block. Although the implementedswitch matrix provides sufficient flexibility in the elec-
trode selection for most experiments, this will be further
improved in the near future by incorporating two routing
switches/memory bits per electrode instead of currently
only one. This can easily be realized by using a CMOS-
technology with a feature size of 0.35 �m or less, while
preserving or even increasing the array spatial resolution.
A block diagram of the overall device is shown inFig. 20 [53], [129]. For readout, the signals are amplified
and filtered using three stages, each including a two-stage,
Miller-compensated amplifier. The gain is programmable
via the digital interface from 1 to 10 000 in 18 steps to
account for the large variation in the signal amplitudes of
the different cell types. The maximal gain of the first stageis 30. This stage also provides a first-order high-pass filter
with adjustable cutoff frequency (0.3–100 Hz). The
bandwidth is limited with the Miller capacitance to either
15 or 50 kHz in the first, and to 3.5 or 14 kHz in the second
stage. The signals are multiplexed after the second stage,
sampled at 20 kSamples/s per channel and digitized with
an 8-b successive-approximation ADC. The data are
transferred off-chip along with the chip-status informationand a cyclic redundancy check (CRC) for transmission
error detection. More information is available in [129].
The stimulation capability is provided through flash
digital-to-analog converters (DACs) and stimulation buf-
fers (Fig. 20). Electrical stimuli can be delivered through
two 10-b DACs, which enable the simultaneous application
of two different stimulation patterns to two selected
subsets of the 126 connected electrodes. The most recentimplementation of the stimulation circuit features voltage
and current stimulation and relies on a set of digitally
controlled switches to switch between these two stimula-
tion modes [144]. The same class-AB operational amplifier
is used for both modes and is operated in high output
current, i.e., voltage mode or high gain, i.e., current mode.
In voltage mode, the circuit is configured as a voltage
follower with low output impedance to approximate anideal voltage source. The class-AB functionality is realized
by a local common-mode feedback structure in the
amplifier [145]. In current mode, the circuit is a positive
current conveyor of type II (CCII+) [146]. The CCII+ is
composed of the same operational amplifier that is used in
Fig. 20. Block diagram showing the electrode array and addressing and the on-chip circuitry. The figures at the top indicate the number of
realizations: 126 readout/stimulation channels, 16 ADCs for readout, 2 DACs for stimulation. Two channels are used to read out the on-chip
temperature sensor and the potential on a separate electrode without amplification or filtering. The digital circuitry, which interfaces with the
outside world, includes a recording control unit and a command decoder. The probe access is for testing purposes. Reprinted with permission
from [129].
Hierlemann et al.: Growing Cells Atop Microelectronic Chips: Interfacing Electrogenic Cells In Vitro
272 Proceedings of the IEEE | Vol. 99, No. 2, February 2011
voltage configuration (now with high gain), plus additional
transistors. This circuit features high output impedance,
since it has to approximate an ideal current source, and it
copies the current from the negative input terminal of the
operational amplifier to the output stage. For more details
on the stimulation circuit, see [129] and [144].
Two additional channels are used to record the on-chip
temperature and the electrode DC potential. A onceprogrammable 16-b ID identifies the chips for automated
information access and transfer to a database. The low-
power design minimizes heating of the biological prepa-
ration [52], [129].
3) Fabrication, Packaging, and Setup: Both MEA chips
have been fabricated in an industrial 0.6-�m CMOS
process with three metal layers, two poly silicon layers,and a high-resistance poly silicon layer at XFAB, Erfurt,
Germany.5 Special post-CMOS processing and packaging
steps are necessary to make the packaged chip ready for
operation under physiological conditions in liquid phase.
The contact material as received from the CMOS
foundry is aluminum, a material, which is known to be
unstable in physiological solutions and to be toxic for many
cells. To avoid cell poisoning and undesirable electro-chemistry or chip corrosion, the electrode material was
selected to be platinum. A two-mask postprocessing proce-
dure was applied to cover the Al electrodes with biocom-
patible platinum. During this processing the electrodes
were shifted sideways [Fig. 21(a)] [147], so that the alu-
minum could be sealed with a passivation stack. The
passivation stack consisted of alternating Si3N4 and SiO2
layers with a total layer thickness of 1.6 �m. The finalelectrode location is on top of the original chip passivation
shifted away from the CMOS metal contacts (see also
Fig. 17). The opening of the passivation stack in the areas
of the new electrode locations leaves platinum metal
electrodes exposed to the liquid. Pt-black was afterwards
grown to reduce the electrode impedance [not shown in
Fig. 21(a)].
Neurons have been cultured for many months on the
chips with this 1.6-�m passivation stack, and the chips
have been reused multiple times. Each of the processed
chips has been mounted and wire-bonded onto a small
printed-circuit board [Fig. 21(b)]. A water-resistant
medical epoxy was used to encapsulate the bond wiresand pads. A glass ring forms a bath that contains a suitable
amount of cell medium.
The recording setup includes a chip-support board, an
FPGA for data acquisition and feedback, and a laptop
computer for data analysis, visualization, and storage
(Fig. 22) [129]. The chip-support board can be placed in
an incubator (humidity contents 60% and even larger) and
provides sockets for five packaged chips [Fig. 21(b)] thatcan be operated simultaneously. This is to avoid mechanical
perturbation by handling devices with plated cells prior to
measurements. The support board provides all necessary
clock and digital control signals, as well as all required
analog references to the chips. The value of the references
can be programmed using the on-board microprocessor.
This processor also monitors temperature and humidity. To
minimize the amount of required connections, the data areserialized on the board and sent via two twisted-pair links at
16 MB/s. By using a serial low-voltage differential signal
(LVDS) protocol, a six-wire ribbon cable is sufficient to
connect the board with the rest of the system.
The main data acquisition and feedback control unit is
realized using the Xilinx Virtex-II-Pro FPGA platform.6
This component features fast data-processing algorithms,
implemented on the programmable logic, and also hoststwo PowerPC cores, both capable of running a Linux ope-
rating system with clock frequencies up to 300 MHz
(Fig. 22) [129]. The data are converted back to a parallel
representation, the data streams from the different chips
5http://www.xfab.com/.
Fig. 21. (a) Scanning electron microscope cross-sectional image of the chip showing the post-CMOS processing needed for biocompatibility
and corrosion protection: platinum processing and shifting of the electrodes to the side on top of the original chip passivation.
(b) Packaged chip on a custom-designed printed circuit board. Reprinted from [52] with permission.
6http://www.xilinx.com/support/documentation/user_guides/ug012.pdf.
Hierlemann et al. : Growing Cells Atop Microelectronic Chips: Interfacing Electrogenic Cells In Vitro
Vol. 99, No. 2, February 2011 | Proceedings of the IEEE 273
are separated, and a CRC is performed. The extracted data
are then distributed via an internal data bus to the data
processing units, e.g., spike-detection and data-compression
units. These units use direct memory access (DMA) to write
the results into the system memory, where they can be
accessed from applications running on the PowerPC. As the
chips produce large amounts of data, which need to be storedfor later processing, a compression block was designed. It is a
lossless compression that encodes the difference between
two data points in time using Rice coding [148] with an
adaptive algorithm to update the compression parameters
depending on the measured signal properties [149]. The
application program running on the PowerPC core is
responsible for the data transport between the FPGA core
and the supervising computer.The third component is the software running on the PC
(Fig. 22) [129]. It consists of a server application and sev-
eral clients connected to the server. As high-density MEAs
have a complex structure with an electrode routing matrix
and a programmable gain, the server implements five chip-
emulator modules. Every time the real chip configuration
is changed, the same changes are applied to the corre-
sponding emulators [129]. The client applications can thenretrieve the currently connected electrodes, gain, and
filter settings from the emulators. All communication links
from the chip to the PC are protected by CRCs in either
direction, which enables error detection and ensures data
integrity.
4) Measurement Results: Dissociated and randomly
seeded cell cultures of electrogenic cells, such as neuronsor cardiac myocytes, have been well established as bio-
logical preparations with MEAs [83], [130], [150]–[152].
Primary neurons are usually isolated from brains of em-
bryonic or postnatal rodents, since these cells exhibit a
high degree of plasticity, a characteristic that is required
for developing neuronal networks in vitro [153]. The ability
to record and stimulate cultured cells has provided unique
experimental setups for studying the cellular activity over
extended periods of time, ranging from hours to weeks
[154], [155].
a) Recording: Recordings from cardiomyocytes, also
under dosing of pharmacologically active agents, and re-
cordings from neuronal cells in rat brain slices at sub-
cellular resolution have been selected from a wealth of
measurements and will be shown here.Cardiomyocytes are responsible for heart contractions
and are the dominant cell type in the normal heart with
respect to volume. Irregularities in heartbeat, due to
cardiac electrical dysfunction, are one of the most frequent
causes of mortality and morbidity in the industrialized
society. Therefore, elaborate in vitro models are needed
that enable to study how potent pharmaceutical agents
affect cardiac electrophysiology. Primary neonatal rat car-diomyocytes can be cultured directly on the chip, they
become electrically active very quickly, and provide com-
parably large signal amplitudes. The recording parameters
included a bandpass filter range between 10 Hz and 5 kHz
at a sampling frequency of 20 kHz. The cardiomyocytes
form a confluent layer on the chip and show spontaneous
activity, a regular beating driven by pacemaker heart cells
in the culture. Recordings of field potentials (synchronousactivity of patches of heart cells) of the spontaneously
beating cells after five days in vitro are shown in Fig. 23(a).
In this example, the cells have been activated by dosing
phenylephrine and beat at a rate of about 10 Hz. The sig-
nals feature amplitudes between 1 and 1.5 mV. In com-
paring the extracellular signal of Fig. 23(a) and (b) with
the intracellular signals displayed in Fig. 4(b), it is obvious
that it is mostly the fast Naþ current contribution to po-tential changes that is recorded by extracellular measure-
ments [61], as the signal duration is on the order of only
1–2 ms.
Drastic changes in the signal characteristics can be ob-
served upon dosing, e.g., lidocaine to the beating culture.
Lidocaine is a common local anesthetic and an antiar-
rhythmic drug. Lidocaine blocks the fast voltage-gated
sodium (Naþ) channels in the cell membrane and alters
Fig. 22. Schematic of the overall recording setup, which includes a chip-support board, an FPGA for data acquisition and feedback, and
a laptop computer for data analysis, visualization, and storage. A photograph of the support board with plugged-in chips is shown on the left-hand
side. �C denotes microcontroller, LVDS is low-voltage differential signal, and UART means universal asynchronous receiver transmitter.
Reprinted with permission from [129].
Hierlemann et al.: Growing Cells Atop Microelectronic Chips: Interfacing Electrogenic Cells In Vitro
274 Proceedings of the IEEE | Vol. 99, No. 2, February 2011
the depolarization of cardiac cells and neurons. With suf-
ficient blockade, the membrane of, e.g., a presynaptic
neuron will not depolarize and so fail to transmit an action
potential, leading to its anesthetic effects. The effect of
lidocaine dosage can be seen in Fig. 23(b) and (c). The use
of a larger array of electrodes enables the recording and
evaluation of a multitude of relevant parameters. The data
in each row represent an average over a large set of cardiacelectrical potentials, recorded with 126 electrodes. The
dosage of lidocaine drastically alters the waveform as can
be seen in the first column when comparing peak-to-peak
voltages and the time span between the positive and
negative peak. As the fast sodium channels are blocked, the
amplitude drastically drops from 2.2 to 0.37 mVPP, and the
time span between the peaks is increased from 0.35 to 2.8 ms.
The wave propagation speed, i.e., the speed, at which theelectrical potential wave travels across the electrode array
area, is also massively reduced from 360 to 105 mm/s,
whereas the culture beating frequency increases from 0.7 to
0.9 Hz. The dependence on the concentration of the dosed
pharmaceutical agent can be seen in comparing the graphs
and values for 5 and 10 mg/ml [Fig. 23(b), second and third
row, and Fig. 23(c)]. All the above mentioned parameters can
be used to characterize the effects of drug dosage on tissuelevel and provide complementary information to single-cell
patch clamp measurements currently routinely conducted in
pharmaceutical industry. For more details on the interpre-
tation of the signal characteristics and waveforms of
extracellularly recorded cardiomyocyte signals, also upon
dosage of pharmacologically important substances, see [61],
[156], and [157]. Further, MEAs have been used to assess the
effects of lentivirus-induced genetic modifications of cardi-
omyocytes and to show the possibilities of using cardiac
microtissues transplanted onto confluent cultures as pace-maker units [158].
As an exemplary neuronal preparation, acute sagittal
cerebellar slices from rats were placed on high-density
CMOS MEAs [Fig. 1(b)], and the temporal evolution of
single-cell action potentials was studied in detail. The signals
of a single Purkinje neuronal cell, the most important cell
type in this kind of preparation, were simultaneously visible
on more than 50 of the tightly spaced electrodes [53].Purkinje cells (PCs) are spontaneously active and are
effectively disconnected from each other because the parallel
fibers are cut in a sagittal slice [48]. In addition, the
excitatory input from the deep cerebellar nuclei via climbing
fibers is missing. The PC electrical fields can thus be
considered independent from each other, which facilitates
spike sorting and eases waveform interpretations.
Spikes in the voltage traces varied considerably inshape, and the signals on neighboring electrodes showed a
high degree of synchronicity, which, respectively, indicates
Fig. 23. (a) Field potential recording of a confluent layer of regularly beating cardiomyocytes from neonatal rat (embryonic day 5) after
phenylephrine dosage (100 �M/L) after five days in vitro. (b) Characteristic changes upon dosage of different concentrations of lidocaine
(5 and 10 mg/ml) to a cardiac culture. The first column shows average signals of all electrical potentials recorded within a 20-s time window.
The raster plots in the second column show the distribution of electrical activity over all electrodes or channels as a function of time
(a rhythmic synchronous activity pattern can be observed). The propagation plots in the third column show the spreading of the electrical wave,
which originates from the upper left corner, with time. The lines labeled with 5, 10, 15, etc., represent the location of the wave front at 5, 10, and
15 ms after activity starts. (c) Characteristic parameters that change upon lidocaine dosage; Upp is the peak-to-peak voltage of the signal,
tPP is the time span between positive and negative peak, v denotes the wave propagation speed, and f is the culture beating frequency.
Hierlemann et al. : Growing Cells Atop Microelectronic Chips: Interfacing Electrogenic Cells In Vitro
Vol. 99, No. 2, February 2011 | Proceedings of the IEEE 275
that different neurons contribute to the overall signal on
each electrode and that the same neuron provides signals
on various electrodes [Fig. 24(a)]. The spike shape distri-butions were analyzed using independent component
analysis (ICA) to identify and separate spikes originating
from different neurons.
Spike-triggered averages of the signals on all electrodes
in a region of interest were calculated for each identified
unit, and the potential distribution generated by an indi-
vidual Purkinje cell was analyzed as shown in Fig. 24(b).
Spike amplitudes ranged from the detection limit (�35 �V,�4� noise level) up to a few hundred �V, with firing rates
of 7–35 Hz. No significant cross correlation of spike trains
from neighboring PCs was observed. This fact and the
large number of electrodes allowed for an effectiveapplication of independent-component analysis (ICA).
The firing rates may be somewhat underestimated, since
priority was given to precision over sensitivity during
spike sorting in order to obtain accurate cell footprints, i.e.,
accurate extracellular-potential distributions.
The high-temporal resolution characteristics of the
evolution of a single action potential in the identified cells
could be demonstrated (Fig. 25). The current-source-density analysis (second spatial derivative of the electrical
Fig. 24. (a) Recordings of spontaneous activity obtained from an acute parasagittal cerebellar slice. Electrical signals from six neighboring
electrodes: Differently shaped spikes identified through spike sorting are marked in color. Each cell is detectable on more than one electrode,
but not on all six. The electrode number assignment is shown at the left. (b) Footprint of a cell reconstructed from recordings at full spatial
resolution: Potential distribution at the time of the first negative peak (circle: center of all negative peaks weighted with the peak amplitudes;
square: center of all positive peaks). Frequency band: 5 Hz–3.5 kHz. Colored contour lines represent equipotential lines at half-peak amplitude,
i.e., �63 �V for the blue line in the negative or somatic region (blue), and þ18 �V for the red line in the positive or dendritic region (red).
Reprinted with permission from [53].
Fig. 25. Current-source-density analysis for the same cell at several points in time (green: current sink; orange: current source). The sink
moves from the soma at 0.4 ms to the proximal dendrites at 0.6 ms and covers the dendritic area, while the soma repolarizes. The outlines of
cell soma (blue) and dendritic area (red) again represent equipotential lines at half-peak amplitude. The gray dots mark the electrodes used
to record from. Reprinted with permission from [159].
Hierlemann et al.: Growing Cells Atop Microelectronic Chips: Interfacing Electrogenic Cells In Vitro
276 Proceedings of the IEEE | Vol. 99, No. 2, February 2011
signals) revealed distinct dipole fields oriented in parallel
to the chip surface. At spike onset, there was a pronounced
current sink around the presumptive location of the PC
soma and an equivalent, but spatially less focused, source
in the region of the dendritic tree (Fig. 25). The sink then
shifted towards the region of the proximal dendrites,
where it slowly faded away. This passive backpropagation
of an action potential, i.e., the gradual movement of thecurrent sink into the region of the proximal dendrites,
where initially the source has been located, was clearly
visible and reproducible.
b) Stimulation: Stimulation features, as also realized
by other groups for in vivo applications [5], [160], [161],
have been implemented on the CMOS chips here as well.
Voltage and current stimulation was included in the
designs, and stimulation buffers have been implemented ateach electrode (Section IV-E1) or in each channel
(Section IV-E2). The implementation of buffers is due to
the fact that the number of activated stimulation
electrodes and, therefore, the load for the stimulation
circuitry can vary. The buffers are powered down, if the
respective electrode or channel has not been selected for
stimulation (see Figs. 18 and 20). In order to stimulate
with a variety of waveforms, a stimulation signal samplingrate of 60 kHz has been chosen, which, in turn, entails the
use of buffers with slew rates of 3 V/16 �s ¼ 0.2 V/�s at a
load of 20 nF.
An important aspect for stimulation includes the elec-
trode impedance. Studies on the electrode impedance have
been reported in [73], [89], and [162], from which it was
evident that the impedance of the electrode varies in de-
pendence of electrode size, electrode material, and cell
growth [163]. The capacitance of a 30-�m diameter elec-
trode has been measured to vary from 200 pF for bright
platinum (Pt) to 20 nF for Pt-black. The charge-transfer
resistance of the Pt electrode is on the order of several
hundred megaohms.
Electrical stimulation pulses affect the recordings on
the stimulating electrode and on neighboring electrodes by
creating large artifacts, which may superimpose on ormask action potentials occurring during the decay time of
an artifact. There are several methods to eliminate stimu-
lation artifacts, which are computationally expensive and/
or require real-time data processing [114], [115]. An alter-
native is to include artifact suppression measures in the
stimulation/recording circuitry [5], [56], [160], [161],
[164]. In the systems presented here, artifact cancellation
has been implemented directly in the pixel/channel cir-cuitry. To prevent high-pass filter saturation upon applica-
tion of large signal amplitudes, a reset switch (see Fig. 18)
has been implemented in the feedback of the first am-
plifier. This switch holds the amplifier in the buffer mode
and brings the filter back into the operation range within
less than 100 �s [104], [105].
The effectiveness of the on-chip reset function for sti-
mulation artifact suppression is demonstrated inFig. 26(a). Bipolar pulses (220-�s overall duration) of
different amplitudes (�0.15 V and�1 V) in saline solution
have been applied, while the reset has been operational
(red trace �0.15 V; green trace �1 V) or not (blue trace
�0.15 V; black trace �1 V). The recordings from the sti-
mulating electrode are shown [105]. The readout filter for
the red and green trace was reset 50 �s before the sti-
mulation pulse was applied, and kept until 50 �s after the
Fig. 26. (a) Efficiency of the on-chip reset function for stimulation. Bipolar pulses (220-�s overall duration) of different amplitudes
(�0.15 V and �1 V) in saline solution have been applied, and the reset has been operational (red trace �0.15 V; green trace �1 V) or not
(blue trace �0.15 V; black trace �1 V). Without reset, it takes 20–100 ms for the recording circuitry to return to the measurement range.
The inset shows a closeup of the initial 20 ms at higher temporal resolution. Reprinted with permission from [105]. (b) Successful excitation
of spikes in cortical neurons from rat brain at 250-�m distance from the stimulation site. A single bipolar stimulation pulse of �800 mV and
50-�s duration was used to stimulate the cells in this example. The inset shows the poststimulus time histogram of this channel based on
142 stimulation pulses. Reprinted with permission from [105].
Hierlemann et al. : Growing Cells Atop Microelectronic Chips: Interfacing Electrogenic Cells In Vitro
Vol. 99, No. 2, February 2011 | Proceedings of the IEEE 277
stimulation pulse was finished. Without reset, it takes
20–100 ms (depending on the stimulation amplitude) for
the recording circuitry to return to the measurementrange. With reset, it takes less than 5 ms for the recording
circuitry to return to the measurement range. It is worth
mentioning that a stimulation sequence always ends with a
stimulation of a value close to the equilibrium potential of
the electrode before simulation so that the electrode
quickly returns to its equilibrium potential after the
stimulation procedure.
Stimulation experiments have been carried out withcardiomyocytes and neuronal cultures [105], [128], [129],
[144], [165]. Fig. 26(b) shows an example of a successful
excitation of spikes in a culture of cortical rat neurons
recorded at a 250-�m distance from the stimulation site
(low-density chip [104], [105]). A single bipolar stimula-
tion pulse of �800 mV and 50-�s duration was used to
stimulate the cells in this example. The inset shows the
poststimulus time histogram of this channel that includesthe results of 142 subsequent stimulation pulses. The first
event or neuronal signal generally occurs between 2 and
3 ms after the stimulation pulse with a very high proba-
bility of 96%.
The switching scheme of the high-density array that is
generally used to route a subset of electrodes to the read-
out channels of the system (Section IV-E2) can also be
used for stimulation. This means that every single elec-trode in the array can be chosen as the stimulation point.
This way, very localized stimulation can be performed
through a single 7-�m diameter electrode, so that generally
high spatio–temporal resolution stimulation and recording
can be performed using the MEA. Local stimulation and
the respective response propagation tracing through the
culture have been performed in an experiment shown in
Fig. 27 [166]. Stimuli (biphasic voltage pulses between
100- and 400-�s duration and þ=�0.7–1.3-V magnitude
per phase) were applied to a single electrode located in thebottom left corner and evoked activity of a subset of neu-
rons. Raw voltage traces are coded in color scale in the first
row of Fig. 27. A somatic action potential (blue) produces
spikes on multiple neighboring electrodes with peaks often
around �400 �V, some even exceeding a millivolt. How-
ever, the scale was reduced here in order to observe pro-
pagating action potentials. These are clearly seen when
plotting the minimum voltage recorded on each electrodesince the stimulus, as done in the bottom row. Each frame
is composed of 108 electrode configurations, and re-
sponses were averaged over 60 trials per configuration.
Each configuration comprises 6 � 17 electrodes and has
been scanned across the entire array. The culture was six
weeks old and seeded with 10 000 cells. As is evident from
the bottom row of Fig. 27, it was possible to observe the
antidromic propagation of action potentials along axonsand the subsequent occurrence of somatic action poten-
tials in the respective cells (blue spots, A and B in upper
row). Obviously, axonal electrical activity of cells A and B
has been triggered at the stimulation site, and the elec-
trical wave has then backpropagated along the axon to the
soma and led to cell depolarization [166]. It is remarkable
that the comparably weak electric activity of axons, tiny
structures of 1-�m diameter can be recorded by meansof high-density arrays of extracellular electrodes.
Finally, it is important to mention that the high-density
chips have been used to perform more than 20 000 stimu-
lations in a rat neuronal culture without damage to the
electrodes or the chips. This opens a route to a variety of
interesting experiments and, at the same time, proves the
efficiency of the passivation and packaging procedure.
Fig. 27. Propagation of axonal action potentials. Raw voltage traces (top row) and traces binned over 2 ms (bottom row) are color-coded.
Stimuli applied to a single electrode (arrow in first panel) evoked a subset of neurons, and somatic action potentials (blue) were detected by
multiple neighboring electrodes. The color scale was reduced in order to zoom-in on axonal APs (yellow). To sample all electrodes, coherent
6 � 17 electrode recording configurations were scanned across the array, and responses after 60 stimuli per configuration were averaged.
Reprinted with permission from [166].
Hierlemann et al.: Growing Cells Atop Microelectronic Chips: Interfacing Electrogenic Cells In Vitro
278 Proceedings of the IEEE | Vol. 99, No. 2, February 2011
V. CONCLUSION AND OUTLOOK
The use of CMOS technology offers decisive advantages for
recording electrical signals from cells in vitro with regardto number and density of recording sites or electrodes that
can be used and with regard to obtained signal quality.
Integrated multiplexers and on-chip electronics enable
high-quality recording or stimulation through a large num-
ber of transducers and allow for attaining cellular or even
subcellular resolution. Moreover, elaborate user interfaces
along with electronic functionality on-chip and chip
self-identification features help the nonexpert user tocomfortably work with these high-performance devices.
A strong thrust goes in the direction to further increase
the number of stimulation and recording electrodes. In
pursuing this goal it has to be taken into account that a
large number of electrodes also generates a huge volume of
recorded data. Therefore, data sorting, standardized or
automated data interpretation, and data compression will
become pivotal issues. Methods like online spike detectionusing, e.g., an FPGA, will be very important to be able to
handle the data volume of long-term recordings from elec-
trode arrays with several hundreds or thousands of elec-
trodes. The approach of having a high-density array of
electrodes, where only a selected subset of the electrodes
in regions of interest is connected to recording and stimu-
lation channels by means of a switch matrix, seems also to
be beneficial for controlling the amount of recorded data.Another important trend includes the integration of other
sensor types into the system, such as chemical sensors
(electrolyte concentration, pH, and oxygen content) to
observe cellular metabolic characteristics or to monitor
neurotransmitters, or impedance sensors to study the local
electrolyte concentration, the cell attachment, or possibly
to get information on cell location and tissue morphology.
CMOS microsystem platforms will have, most proba-bly, a major scientific and technological impact, particu-
larly in neuroscience and neuromedical research. They
will enable fundamental neurophysiological insights at the
cellular and circuit level for investigating dynamic altera-
tions in neuronal connectivity patterns and provide a new
approach to investigate the behavior and information pro-cessing capabilities (including learning) of large popula-
tions of neurons in vitro or in slices. They may even open
routes to the design of dedicated hybrid Bintelligent[systems combining evolutionarily optimized highly paral-
lel information processing capabilities (of, e.g., brain cells)
with the more serial ones of microelectronics.
The CMOS-based systems will also find applications in
medical diagnostics and pharmacology. The systems can beused for assessing dose-related effects of all kinds of
chemicals or pharmacologically active substances [25],
[45], [46], [167]. This holds particularly true for long-term
measurements and multiple dosings or dosing sequences,
which cannot be studied using the patch clamp method
owing to the inherent short cell viability time (usually on
the order of hours) as a consequence of the invasive nature
of the method.Finally, CMOS-based integrated systems may also have
a large impact on in vivo applications in the field of neural
probes and prosthetics [4], [5], [50], [168]–[171]. There,
they may constitute a viable approach to recover skeletal
muscle function [168], [169] and to restore visual per-
ception [172], [173] or functions of the auditory system
[174], [175]. For practical neural prostheses, viable inter-
faces with tissue for months, years, or even decades will berequired. Mismatch in the mechanical properties (stiff
devices and soft tissue), inflammations, depletion of
neurons in the insertion region, tissue reaction, and probe
encapsulation with glial tissue yet compromise the long-
term electrical-recording quality and reliability and will
have to be addressed [50]. h
Acknowledgment
The authors would like to thank R. Pedron, K.-U. Kirstein,D. Scheiwiller, M. Ballini, J. Mueller, W. Franks, and
F. Greve for contributing to the development of the overall
system. U. Wahlen, D. Jackel, J. Sedivy, C. D. Bustamante,
D. Bakkum, and B. Roscic are acknowledged for help with
the measurements and the data analysis.
RE FERENCES
[1] E. Neher and B. Sakmann, BSingle-channelcurrents recorded from membrane ofdenervated frog muscle fibres,[ Nature,vol. 260, pp. 799–802, Apr. 29, 1976.
[2] K. Cole, BDynamic electrical characteristicsof the squid axon membrane,[ Arch. Sci.Physiol., vol. 3, pp. 253–258, 1949.
[3] P. Fromherz, BElectrical interfacing ofnerve cells and semiconductor chips,[Chemphyschem, vol. 3, pp. 276–284,Mar. 12, 2002.
[4] K. D. Wise, D. J. Anderson, J. F. Hetke,D. R. Kipke, and K. Najafi, BWirelessimplantable microsystems: High-densityelectronic interfaces to the nervous system,[Proc. IEEE, vol. 92, no. 1, pp. 76–97,Jan. 2004.
[5] R. H. Olsson, III, D. L. Buhl, A. M. Sirota,G. Buzsaki, and K. D. Wise, BBand-tunable
and multiplexed integrated circuits forsimultaneous recording and stimulationwith microelectrode arrays,[ IEEE Trans.Biomed. Eng., vol. 52, no. 7, pp. 1303–1311,Jul. 2005.
[6] A. G. Kleber and Y. Rudy, BBasicmechanisms of cardiac impulse propagationand associated arrhythmias,[ Physiol. Rev.,vol. 84, pp. 431–488, 2003.
[7] W. L. C. Rutten, BSelective electricalinterfaces with the nervous system,[ Annu.Rev. Biomed. Eng., vol. 4, pp. 407–452, 2002.
[8] Y. Jimbo and H. P. Robinson, BPropagationof spontaneous synchronized activity incortical slice cultures recorded by planarelectrode arrays,[ Bioelectrochemistry,vol. 51, pp. 107–115, 2000.
[9] J. van Pelt, P. S. Wolters, M. A. Corner,W. L. C. Rutten, and G. J. A. Ramakers,BLong-term characterization of firing
dynamics of spontaneous bursts in culturedneural networks,[ IEEE Trans. Biomed.Eng., vol. 51, no. 11, pp. 2051–2062,Nov. 2004.
[10] S. I. Morefield, E. W. Keefer,K. D. Chapman, and G. W. Gross,BDrug evaluations using neuronal networkscultured on microelectrode arrays,[Biosens. Bioelectron., vol. 15, pp. 383–396,Oct. 2000.
[11] J. J. Pancrazio, S. A. Graya, Y. S. Shubinb,N. Kulaginaa, D. S. Cuttinoc, K. M. Shaffera,K. Eisemanna, A. Currand, B. Zimd,G. W. Grossd, and T. J. O’Shaughnessy,BA portable microelectrode array recordingsystem incorporating cultured neuronalnetworks for neurotoxin detection,[Biosens. Bioelectron., vol. 18, pp. 1339–1347,Oct. 1, 2003.
[12] B. D. DeBusschere and G. T. A. Kovacs,BPortable cell-based biosensor system
Hierlemann et al. : Growing Cells Atop Microelectronic Chips: Interfacing Electrogenic Cells In Vitro
Vol. 99, No. 2, February 2011 | Proceedings of the IEEE 279
using integrated CMOS cell-cartridges,[Biosens. Bioelectron., vol. 16, pp. 543–556,Sep. 2001.
[13] M. Taketani, Advances in NetworkElectrophysiology Using Multi-ElectrodeArrays. New York: Springer-Verlag,2006.
[14] G. T. A. Kovacs, BElectronic sensorswith living cellular components,[ Proc. IEEE,vol. 91, no. 6, pp. 915–929, Jun. 2003.
[15] B. J. Baker, E. K. Kosmidis, D. Vucinic,C. X. Falk, L. B. Cohen, M. Djurisic, andD. Zecevic, BImaging brain activity withvoltage- and calcium-sensitive dyes,[Cell. Mol. Neurobiol., vol. 25, pp. 245–282,Mar. 2005.
[16] Z. A. Peterlin, J. Kozloski, B.-Q. Mao,A. Tsiola, and R. Yuste, BOpticalprobing of neuronal circuits with calciumindicators,[ Proc. Nat. Acad. Sci. USA, vol. 97,pp. 3619–3624, Mar. 28, 2000.
[17] A. L. Obaid, L. M. Loew,J. P. Wuskell, and B. M. Salzberg, BNovelnaphthylstyryl-pyridium potentiometricdyes offer advantages for neural networkanalysis,[ J. Neurosci. Methods, vol. 134,pp. 179–190, Apr. 30, 2004.
[18] M. Voelker and P. Fromherz, BSignaltransmission from individual mammaliannerve cell to field-effect transistor,[Small, vol. 1, pp. 206–210, 2005.
[19] M. Jenkner, B. Muller, and P. Fromherz,BInterfacing a silicon chip to pairs of snailneurons connected by electrical synapses,[Biol. Cybern., vol. 84, pp. 239–249,Apr. 2001.
[20] P. Bergveld, BDevelopment of anion-sensitive solid-state device forneurophysiological measurements,[IEEE Trans. Biomed. Eng., vol. BME-17,no. 1, pp. 70–71, Jan. 1970.
[21] P. Bergveld, BDevelopment, operation, andapplication of the ion-sensitive field-effecttransistor as a tool for electrophysiology,[IEEE Trans. Biomed. Eng., vol. BME-19, no. 5,pp. 342–351, Sep. 1972.
[22] P. Bergveld, BThirty years of ISFETOLOGYwhat happened in the past 30 years andwhat may happen in the next 30 years,[Sens. Actuators B, Chem., vol. 88, pp. 1–20,2003.
[23] M. P. Maher, J. Pine, J. Wright, andY.-C. Tai, BThe neurochip: A newmultielectrode device for stimulatingand recording from cultured neurons,[J. Neurosci. Methods, vol. 87, pp. 45–56,1999.
[24] S. Martinoia, P. Massobrio, M. Bove, andG. Massobrio, BCultured neurons coupledto microelectrode arrays: Circuit models,simulations and experimental data,[IEEE Trans. Biomed. Eng., vol. 51, no. 5,pp. 859–864, May 2004.
[25] A. Stett, U. Egert, E. Guenther, F. Hofmann,T. Meyer, W. Nisch, and H. Haemmerle,BBiological application of microelectrodearrays in drug discovery and basic research,[Anal. Bioanal. Chem., vol. V377, pp. 486–495,Oct. 2003.
[26] M. O. Heuschkel, M. Fejtl,M. Raggenbass, D. Bertrand, and P. Renaud,BA three-dimensional multi-electrode arrayfor multi-site stimulation and recordingin acute brain slices,[ J. Neurosci. Methods,vol. 114, pp. 135–148, Mar. 2002.
[27] Y. Jimbo, N. Kasai, K. Torimitsu, T. Tateno,and H. P. C. Robinson, BA system forMEA-based multisite stimulation,[IEEE Trans. Biomed. Eng., vol. 50, no. 2,pp. 241–248, Feb. 2003.
[28] G. W. Gross, BSimultaneous single unitrecording in vitro with a photoetchedlaser deinsulated gold multimicroelectrodesurface,[ IEEE Trans. Biomed. Eng.,vol. BME-26, no. 5, pp. 273–279,May 1979.
[29] G. W. Gross, E. Rieske, G. W. Kreutzberg,and A. Meyer, BA new fixed-arraymulti-microelectrode system designedfor long-term monitoring of extracellularsingle unit neuronal activity in vitro,[Neurosci. Lett., vol. 6, pp. 101–105,Nov. 1977.
[30] C. A. Thomas, P. A. Springer, G. E. Loeb,Y. Berwald-Netter, and L. M. Okun,BA miniature microelectrode array tomonitor the bioelectric activity ofcultured cells,[ Experiment. Cell Res.,vol. 74, pp. 61–66, 1972.
[31] K. D. Wise and J. B. Angeli, BAmicroprobe with integrated amplifiersfor neurophysiology,[ in Proc. IEEE Int.Solid-State Circuits Conf., 1971, pp. 100–101.
[32] I. Schoen and P. Fromherz, BThe mechanismof extracellular stimulation of nerve cellson an electrolyte-oxide-semiconductorcapacitor,[ Biophys. J., vol. 92, pp. 1096–1111,2007.
[33] I. Schoen and P. Fromherz, BExtracellularstimulation of mammalian neurons throughrepetitive activation of Na+ channels byweak capacitive currents on a silicon chip,[J. Neurophysiol., vol. 100, pp. 346–357, 2008.
[34] F. Wallrapp and P. Fromherz, BTiO2and HfO2 in electrolyte-oxide-siliconconfiguration for applications inbioelectronics,[ J. Appl. Phys., vol. 99,2006, article 114103.
[35] U. Frey, U. Egert, D. Jackel, J. Sedivy,M. Ballini, P. Livi, F. Faraci, F. Heer,S. Hafizovic, B. Roscic, and A. Hierlemann,BDepth recording capabilities of planar high-density microelectrode arrays,[ in Proc. 4thInt. IEEE EMBS Conf. Neural Eng., Antalya,Turkey, 2009, pp. 207–210.
[36] F. Greve, S. Frerker, A. Greet Bittermann,C. Burkhardt, A. Hierlemann, and H. Hall,BMolecular design and characterization ofthe neuron-microelectrode array interface,[Biomaterials, vol. 28, pp. 5246–5258,Dec. 2007.
[37] W. V. Soussou, G. J. Yoon, R. D. Brinton, andT. W. Berger, BNeuronal networkmorphology and electrophysiology ofhippocampal neurons cultured onsurface-treated multielectrode arrays,[IEEE Trans. Biomed. Eng., vol. 54, no. 7,pp. 1309–1320, Jul. 2007.
[38] U. Egert, B. Schlosshauer, S. Fennrich,W. Nisch, M. Fejtl, T. Knott, T. Muller, andH. Hammerle, BA novel organotypiclong-term culture of the rat hippocampuson substrate-integrated multielectrodearrays,[ Brain Res. Protocols, vol. 2,pp. 229–242, Jun. 1998.
[39] S. A. Boppart, B. C. Wheeler, andC. S. Wallace, BA flexible perforatedmicroelectrode array for extended neuralrecordings,[ IEEE Trans. Biomed. Eng.,vol. 39, no. 1, pp. 37–42, Jan. 1992.
[40] Y. Jimbo, T. Tateno, and H. P. Robinson,BSimultaneous induction of pathway-specificpotentiation and depression in networksof cortical neurons,[ Biophys. J., vol. 76,pp. 670–678, Feb. 1999.
[41] S. Marom and G. Shahaf, BDevelopment,learning and memory in large randomnetworks of cortical neurons: Lessonsbeyond anatomy,[ Quart. Rev. Biophys.,vol. 35, pp. 63–87, Feb. 2002.
[42] G. Shahaf and S. Marom, BLearning innetworks of cortical neurons,[ J. Neurosci.,vol. 21, pp. 8782–8788, Nov. 15, 2001.
[43] M. E. Ruaro, P. Bonifazi, and V. Torre,BTowards the neurocomputer: Imageprocessing and pattern recognition withneuronal cultures,[ IEEE Trans. Biomed. Eng.,vol. 52, no. 3, pp. 371–383, Mar. 2005.
[44] D. A. Wagenaar, Z. Nadasdy, andS. M. Potter, BPersistent dynamic attractorsin activity patterns of cultured neuronalnetworks,[ Phys. Rev. E, Stat. Phys. PlasmasFluids Relat. Interdiscip. Top., vol. 73, 051907,May 2006.
[45] G. W. Gross, B. K. Rhoades, H. M. E. Azzazy,and M.-C. Wu, BThe use of neuronalnetworks on multielectrode arrays asbiosensors,[ Biosens. Bioelectron., vol. 10,pp. 553–567, 1995.
[46] J. J. Pancrazio, P. Bey, Jr., D. Cuttino,J. Kusel, D. Borkholder, K. Shaffer,G. Kovacs, and D. Stenger, BPortablecell-based biosensor system for toxindetection,[ Sens. Actuators B, Chem.,vol. B53, pp. 179–185, 1998.
[47] T. Meyer, C. Leisgen, B. Gonser, andE. Gunther, BQT-screen: High-throughputcardiac safety pharmacology by extracellularelectrophysiology on primary cardiacmyocytes,[ ASSAY Drug Develop. Technol.,vol. 2, pp. 507–515, 2004.
[48] U. Egert, D. Heck, and A. Aertsen,BTwo-dimensional monitoring of spikingnetworks in acute brain slices,[ Exp. BrainRes., vol. 142, pp. 268–274, Jan. 2002.
[49] M. Hutzler, A. Lambacher, B. Eversmann,M. Jenkner, R. Thewes, and P. Fromherz,BHigh-resolution multitransistor arrayrecording of electrical field potentials incultured brain slices,[ J. Neurophysiol.,vol. 96, pp. 1638–1645, 2006.
[50] K. D. Wise, A. M. Sodagar, Y. Ying,M. N. Gulari, G. E. Perlin, and K. Najafi,BMicroelectrodes, microelectronics, andimplantable neural microsystems,[ Proc.IEEE, vol. 96, no. 7, pp. 1184–1202,Jul. 2008.
[51] B. Eversmann, M. Jenkner, F. Hofmann,C. Paulus, R. Brederlow, B. Holzapfl,P. Fromherz, M. Merz, M. Brenner,M. Schreiter, R. Gabl, K. Plehnert,M. Steinhauser, G. Eckstein,D. Schmitt-Landsiedel, and R. Thewes,BA 128� 128 CMOS biosensor array forextracellular recording of neural activity,[IEEE J. Solid-State Circuits, vol. 38, no. 12,pp. 2306–2317, Dec. 2003.
[52] U. Frey, F. Heer, R. Pedron, S. Hafizovic,F. Greve, J. Sedivy, K.-U. Kirstein, andA. Hierlemann, BAn 11 k-electrode126-channel high-density microelectrodearray to interact with electrogenic cells,[ inDig. Tech. Papers IEEE Int. Solid-State CircuitsConf., San Francisco, CA, 2007, pp. 158–593.
[53] U. Frey, U. Egert, F. Heer, S. Hafizovic, andA. Hierlemann, BMicroelectronic systemfor high-resolution mapping of extracellularelectric fields applied to brain slices,[Biosens. Bioelectron., vol. 24, pp. 2191–2198,2009.
[54] L. Berdondini, K. Imfeld, A. Maccione,M. Tedesco, S. Neukom, M. Koudelka-Hep,and S. Martinoia, BActive pixel sensorarray for high spatio-temporal resolutionelectrophysiological recordings from singlecell to large scale neuronal networks,[Lab Chip, vol. 9, pp. 2644–2651,Sep. 21, 2009.
[55] D. Plenz and A. Aertsen, BCurrent sourcedensity profiles of optical recording maps:
Hierlemann et al.: Growing Cells Atop Microelectronic Chips: Interfacing Electrogenic Cells In Vitro
280 Proceedings of the IEEE | Vol. 99, No. 2, February 2011
A new approach to the analysis ofspatio-temporal neural activity patterns,[Eur. J. Neurosci., vol. 5, pp. 437–448, 1993.
[56] P. Hottowy, W. Dabrowski, A. Skoczec, andP. Wiacek, BAn integrated multichannelwaveform generator for large-scalespatio-temporal stimulation of neuraltissue,[ Analog Integr. Circuits SignalProcess., vol. 55, pp. 239–248, Jun. 2008.
[57] C. Sekirnjak, P. Hottowy, A. Sher,W. Dabrowski, A. M. Litke, andE. J. Chichilnisky, BHigh-resolutionelectrical stimulation of primate retinafor epiretinal implant design,[ J. Neurosci.,vol. 28, pp. 4446–4456, Apr. 2008.
[58] H. Lodish, A. Berk, C. A. Kaiser, andM. Krieger, Molecular Cell Biology.New York: W. H. Freeman, 2000.
[59] D. Johnston and S. Wu, Foundations ofCellular Neurophysiology. Cambridge,MA: The MIT Press, 1995.
[60] E. R. Kandel, J. H. Schwartz, andT. M. Jessell, Principles of Neural Science,4th ed. New York: McGraw-Hill, 2000.
[61] M. Halbach, U. Egert, J. Hescheler, andK. Banach, BEstimation of action potentialchanges from field potential recordingsin multicellular mouse cardiac myocytecultures,[ Cell. Physiol. Biochem., vol. 13,pp. 271–284, 2003.
[62] A. L. Hodgkin and A. F. Huxley,BA quantitative description of membranecurrent and its application to conduction andexcitation in nerve,[ J. Physiol., vol. 117,pp. 500–544, 1952.
[63] C. Gold, D. A. Henze, C. Koch, andG. Buzsaki, BOn the origin of theextracellular action potential waveform:A modeling study,[ J. Neurophysiol.,vol. 95, pp. 3113–3128, 2006.
[64] D. A. Henze, Z. Borhegyi, J. Csicsvari,A. Mamiya, K. D. Harris, and G. Buzsaki,BIntracellular features predicted byextracellular recordings in thehippocampus in vivo,[ J. Neurophysiol.,vol. 84, pp. 390–400, 2000.
[65] J. R. Buitenweg, W. L. C. Rutten, andE. Marani, BGeometry-based finite-elementmodeling of the electrical contact betweena cultured neuron and a microelectrode,[IEEE Trans. Biomed. Eng., vol. 50, no. 4,pp. 501–509, Apr. 2003.
[66] J. R. Buitenweg, W. L. C. Rutten, andE. Marani, BModeled channel distributionsexplain extracellular recordings fromcultured neurons sealed to microelectrodes,[IEEE Trans. Biomed. Eng., vol. 49, no. 12, pt. 2,pp. 1580–1590, Dec. 2002.
[67] P. Fromherz, BElectrical interfacing ofnerve cells and semiconductor chips,[Chemphyschem, vol. 3, pp. 276–284, 2002.
[68] A. Lambacher, M. Jenkner, M. Merz,B. Eversmann, R. A. Kaul, F. Hofmann,R. Thewes, and P. Fromherz, BElectricalimaging of neuronal activity bymulti-transistor-array (MTA) recording at7.8 mu m resolution,[ Appl. Phys. A, Mater.Sci. Process., vol. 79, pp. 1607–1611, 2004.
[69] M. Voelker and P. Fromherz, BSignaltransmission from individual mammaliannerve cell to field-effect transistor,[Small J., vol. 1, pp. 1–4, 2005.
[70] M. Jenkner, M. Tartagni, A. Hierlemann,and R. Thewes, BCell-based CMOS sensorand actuator arrays,[ IEEE J. Solid StateCircuits, vol. 39, no. 12, pp. 2431–2437,Dec. 2004.
[71] K. D. Wise and J. B. Angeli, BAmicroprobe with integrated amplifiers for
neurophysiology,[ in Proc. Int. Solid-StateCircuits Conf., 1971, pp. 100–101.
[72] R. C. Gesteland, B. Howland, J. Y. Lettvin,and W. H. Pitts, BComments onmicroelectrodes,[ Proc. IRE, vol. 47, no. 11,pp. 1856–1862, Nov. 1959.
[73] W. Franks, I. Schenker, P. Schmutz, andA. Hierlemann, BImpedance characterizationand modeling of electrodes for biomedicalapplications,[ IEEE Trans. Biomed. Eng.,vol. 52, no. 7, pp. 1295–1302, Jul. 2005,2005.
[74] B. Besl and P. Fromherz, BTransistor arraywith an organotypic brain slice: Fieldpotential records and synaptic currents,[Eur. J. Neurosci., vol. 15, pp. 999–1005,2002.
[75] P. Bonifazi and P. Fromherz, BSilicon chipfor electronic communication betweennerve cells by non-invasive interfacingand analog-digital processing,[ Adv. Mater.,vol. 14, pp. 1190–1193, 2002.
[76] M. Jenkner, B. Muller, and P. Fromherz,BInterfacing a silicon chip to pairs of snailneurons connected by electrical synapses,[Biol. Cybern., vol. 84, pp. 239–249,Apr. 2001.
[77] S. Vassanelli and P. Fromherz, BTransistorrecords of rat hippocampal neurons,[Eur. J. Neurosci., vol. 10, suppl. 10, p. 72,1998.
[78] G. Zeck and P. Fromherz, BNoninvasiveneuroelectronic interfacing with synapticallyconnected snail neurons immobilized on asemiconductor chip,[ Proc. Nat. Acad. Sci.,vol. 98, pp. 10457–10462, 2001.
[79] G. W. Gross, W. Y. Wen, and J. W. Lin,BTransparent indium tin oxide electrodepatterns for extracellular, multisite recordingin neuronal cultures,[ J. Neurosci. Methods,vol. 15, pp. 243–252, Nov./Dec. 1985.
[80] V. Bucher, M. Graf, M. Stelzle, andW. Nisch, BLow-impedance thin-filmpolycrystalline silicon microelectrodes forextracellular stimulation and recording,[Biosens. Bioelectron., vol. 14, pp. 639–649,Oct. 1999.
[81] Y. Jimbo, N. Kasai, K. Torimitsu, T. Tateno,and H. P. C. Robinson, BA system forMEA-based multisite stimulation,[IEEE Trans. Biomed. Eng., vol. 50, no. 2,pp. 241–248, Feb. 2003.
[82] Y. Jimbo, A. Kawana, P. Parodi, and V. Torre,BThe dynamics of a neuronal culture ofdissociated cortical neurons of neonatalrats,[ Biol. Cybern., vol. 83, pp. 1–20,Jul. 2000.
[83] D. A. Wagenaar, R. Madhavan, J. Pine, andS. M. Potter, BControlling bursting incortical cultures with closed-loopmulti-electrode stimulation,[ J. Neurosci.,vol. 25, pp. 680–688, Jan. 19, 2005.
[84] M. Chiappalone, A. Vato, M. (B.) Tedesco,M. Marcoli, F. Davide, and S. Martinoia,BNetworks of neurons coupled tomicroelectrode arrays: A neuronal sensorysystem for pharmacological applications,[Biosens. Bioelectron., vol. 18, pp. 627–634,May 2003.
[85] A. J. Bard and L. R. Faulkner, ElectrochemicalMethods Fundamentals and Applications,2nd ed. New York: Wiley, 2001.
[86] R. W. De Boer and A. Van Oosterom,BElectrical properties of platinum electrodes:Impedance measurements and time-domainanalysis,[ Med. Biol. Eng. Comput., vol. 16,pp. 1–10, 1978.
[87] B. Onaral and H. P. Schwan, BLinearand nonlinear properties of platinum
electrode polarisation. III. Equivalence offrequency- and time-domain behaviour,[Med. Biol. Eng. Comput., vol. 23, pp. 28–32,1985.
[88] E. Barsoukov, Impedance Spectroscopy Theory,Experiment, and Applications, 2nd ed.Hoboken, NJ: Wiley-Interscience, 2005.
[89] F. Heer, W. Franks, A. Blau, S. Taschini,C. Ziegler, A. Hierlemann, and H. Baltes,BCMOS microelectrode array for themonitoring of electrogenic cells,[Biosens. Bioelectron., vol. 20, pp. 358–366,Sep. 15, 2004.
[90] J. J. Pancrazio, J. P. Whelan,D. A. Borkholder, W. Ma, and D. A. Stenger,BDevelopment and application of cell-basedbiosensors,[ Ann. Biomed. Eng., vol. 27,pp. 687–711, 1999.
[91] L. Berdondini, P. D. van der Wal, O. Guenat,N. F. de Rooij, M. Koudelka-Hep, P. Seitz,R. Kaufmann, P. Metzler, N. Blanc, andS. Rohr, BHigh-density electrode arrayfor imaging in vitro electrophysiologicalactivity,[ Biosens. Bioelectron., vol. 21,pp. 167–174, 2005.
[92] M. Janders, U. Egert, M. Stelzle, andW. Nisch, BNovel thin film titanium nitridemicro-electrodes with excellent chargetransfer capability for cell stimulation andsensing applications,[ in Proc. 18th Annu.Int. Conf. IEEE Eng. Med. Biol. Soc., vol. 18,H. Boom, Ed., 1997, vol. 18, pp. 245–247.
[93] J. D. Weiland, D. J. Anderson, andM. S. Humayun, BIn vitro electricalproperties for iridium oxide versus titaniumnitride stimulating electrodes,[ IEEETrans. Biomed. Eng., vol. 49, no. 12, pt. 2,pp. 1574–1579, Dec. 2002.
[94] E. Ben-Jacob and Y. Hanein, BCarbonnanotube micro-electrodes for neuronalinterfacing,[ J. Mater. Chem., vol. 18,pp. 5181–5186, 2008.
[95] T. Gabay, M. Ben-David, I. Kalifa, R. Sorkin,Z. R. Abrams, E. Ben-Jacob, and Y. Hanein,BElectro-chemical and biological propertiesof carbon nanotube based multi-electrodearrays,[ Nanotechnology, vol. 18, Jan. 2007,DOI: 10.1088/0957-4484/18/3/035201.
[96] M. Shein, A. Greenbaum, T. Gabay,R. Sorkin, M. David-Pur, E. Ben-Jacob, andY. Hanein, BEngineered neuronal circuitsshaped and interfaced with carbon nanotubemicroelectrode arrays,[ Biomed. Microdevices,vol. 11, pp. 495–501, Apr. 2009.
[97] G. Massobrio, P. Massobrio, andS. Martinoia, BModeling the neuron-carbonnanotube-ISFET junction to investigatethe electrophysiological neuronal activity,[Nano Lett., vol. 8, pp. 4433–4440, Dec. 2008.
[98] A. Hai, J. Shappir, and M. E. Spira,BIn-cell recordings by extracellularmicroelectrodes,[ Nature Methods, vol. 7,pp. 200–202, Mar. 2010.
[99] R. Huys, D. Braeken, B. Van Meerbergen,K. Winters, W. Eberle, J. Loo, D. Tsvetanova,C. Chen, S. Severi, S. Yitzchaik, M. Spira,J. Shappir, G. Callewaert, G. Borghs, andC. Bartic, BNovel concepts for improvedcommunication between nerve cells andsilicon electronic devices,[ Solid-StateElectron., vol. 52, pp. 533–539, 2008.
[100] S. R. Chang, C. H. Chang, J. S. Lin,S. C. Lu, Y. T. Lee, S. R. Yeh, and H. Chen,BDie-level, post-CMOS processes forfabricating open-gate, field-effect biosensorarrays with on-chip circuitry,[ J. Micromech.Microeng., vol. 18, 115032, 2008.
[101] S. R. Chang and H. Chen, BACMOS-compatible, low-noise ISFETbased on high efficiency ion-modulated
Hierlemann et al. : Growing Cells Atop Microelectronic Chips: Interfacing Electrogenic Cells In Vitro
Vol. 99, No. 2, February 2011 | Proceedings of the IEEE 281
lateral-bipolar conduction,[ Sensors,vol. 9, pp. 8336–8348, 2009.
[102] C. Cane, I. Gracia, and A. Merlos,BMicrotechnologies for pH ISFETchemical sensors,[ Microelectron. J.,vol. 28, pp. 389–405, 1997.
[103] F. Hofmann, B. Eversmann, M. Jenkner,A. Frey, M. Merz, T. Birkenmaier,P. Fromherz, M. Schreiter, R. Gabl,K. Plehnert, M. Steinhauser, G. Eckstein,and R. Thewes, BTechnology aspects ofa CMOS neuro-sensor: Back end processand packaging,[ in Proc. Eur. Conf.Solid-State Device Res., Estoril, Portugal,2003, pp. 167–170.
[104] F. Heer, S. Hafizovic, W. Franks, A. Blau,C. Ziegler, and A. Hierlemann, BCMOSmicroelectrode array for bidirectionalinteraction with neuronal networks,[IEEE J. Solid-State Circuits, vol. 41, no. 7,pp. 1620–1629, Jul. 2006.
[105] F. Heer, S. Hafizovic, T. Ugniwenko,U. Frey, W. Franks, E. Perriard,J.-C. Perriard, A. Blau, C. Ziegler, andA. Hierlemann, BSingle-chip microelectronicsystem to interface with living cells,[Biosens. Bioelectron., vol. 22, pp. 2546–2553,May 2007.
[106] D. J. Bakkum, Z. C. Chao, and S. M. Potter,BSpatio-temporal electrical stimuli shapebehavior of an embodied cortical network ina goal-directed learning task,[ J. Neural Eng.,vol. 5, pp. 310–323, Sep. 2008.
[107] D. A. Wagenaar and S. M. Potter,BA versatile all-channel stimulator forelectrode arrays, with real-time control,[J. Neural Eng., vol. 1, pp. 39–45, Mar. 2004.
[108] P. Bonifazi and P. Fromherz, BSilicon chipfor electronic communication betweennerve cells by non-invasive interfacing andanalog-digital processing,[ Adv. Mater.,vol. 14, pp. 1190–1193, Sep. 3, 2002.
[109] G. W. Gross, B. K. Rhoades, D. L. Reust, andF. U. Schwalm, BStimulation of monolayernetworks in culture through thin-filmindium-tin oxide recording electrodes,[J. Neurosci. Methods, vol. 50, pp. 131–143,Nov. 1993.
[110] J. R. Buitenweg, W. L. C. Rutten, andE. Marani, BExtracellular stimulationwindow explained by a geometry-basedmodel of the neuron-electrode contact,[IEEE Trans. Biomed. Eng., vol. 49, no. 12, pt. 2,pp. 1591–1599, Dec. 2002.
[111] C. C. McIntyre and W. M. Grill,BExtracellular stimulation of centralneurons: Influence of stimulus waveformand frequency on neuronal output,[J. Neurophysiol., vol. 88, pp. 1592–1604,Oct. 2002.
[112] Y. Jimbo and A. Kawana,BElectrical-stimulation and recording fromcultured neurons using a planar electrodearray,[ Bioelectrochem. Bioenergetics,vol. 29, pp. 193–204, Dec. 1992.
[113] D. A. Wagenaar, J. Pine, and S. M. Potter,BEffective parameters for stimulation ofdissociated cultures using multi-electrodearrays,[ J. Neurosci. Methods, vol. 138,pp. 27–37, Sep. 2004.
[114] J. W. Gnadt, S. D. Echols, A. Yildirim,H. Zhang, and K. Paul, BSpectral cancellationof microstimulation artifact for simultaneousneural recording in situ,[ IEEE Trans.Biomed. Eng., vol. 50, no. 10, pp. 1129–1135,Oct. 2003.
[115] D. A. Wagenaar and S. M. Potter, BReal-timemulti-channel stimulus artifact suppressionby local curve fitting,[ J. Neurosci. Methods,vol. 120, pp. 113–120, Oct. 30, 2002.
[116] D. B. McCreery, W. F. Agnew, T. G. H. Yuen,and L. Bullara, BCharge-density and chargeper phase as cofactors in neural injuryinduced by electrical-stimulation,[IEEE Trans. Biomed. Eng., vol. 37, no. 10,pp. 996–1001, Oct. 1990.
[117] D. A. Wagenaar, R. Madhavan, J. Pine, andS. M. Potter, BControlling bursting incortical cultures with closed-loopmulti-electrode stimulation,[ J. Neurosci.,vol. 25, pp. 680–688, 2005.
[118] D. A. Wagenaar and S. M. Potter, BReal-timemulti-channel stimulus artifact suppressionby local curve fitting,[ J. Neurosci. Methods,vol. 120, pp. 113–120, 2002.
[119] D. A. Wagenaar and S. M. Potter, BAversatile all-channel stimulator for electrodearrays, with real-time control,[ J. NeuralEng., vol. 1, pp. 39–45, 2004.
[120] M. Merz and P. Fromherz, BSilicon chipinterfaced with a geometrically defined netof snail neurons,[ Adv. Funct. Mater., vol. 15,pp. 739–744, May 2005.
[121] C. Sprossler, M. Denyer, S. Britland,W. Knoll, and A. Offenhausser, BElectricalrecordings from rat cardiac muscle cellsusing field-effect transistors,[ Phys. Rev. E,Stat. Phys. Plasmas Fluids Relat. Interdiscip.Top., vol. 60, pp. 2171–2176, 1999.
[122] A. Offenhausser and W. Knoll,BCell-transistor hybrid systems and theirpotential applications,[ Trends Biotechnol.,vol. 19, pp. 62–66, Feb. 2001.
[123] W. Baumann, E. Schreiber, G. Krause,A. Podssun, S. Homma, H. Anlauf, I. Freund,and M. Lehmann, BMultiparametricneurosensor microchip,[ in Proc.Eurosensors XVI, 2002, pt. 3, p. 685.
[124] K. Imfeld, A. Garenne, S. Neukom,A. Maccione, S. Martinoia,M. Koudelka-Hep, and L. Berdondini,BHigh-resolution MEA platform for in-vitroelectrogenic cell networks imaging,[ in Proc.29th Annu. Int. Conf. IEEE Eng. Med. Biol. Soc.,Lyon, France, 2007, pp. 6085–6088.
[125] A. Maccione, F. Benfenati, L. Berdondini,M. Gandolfo, M Tedesco, S. Martinoia,K. Imfeld, and M. Koudelka-Hep, BImagingextracellular neuronal signaling on highresolution microelectrode arrays (MEAs)Hippocampal cultures coupled with a highresolution neuroelectronic interface,[ inProc. 4th Int. IEEE EMBS Conf. Neural Eng.,Antalya, Turkey, 2009, pp. 234–237.
[126] G. T. A. Kovacs, BElectronic sensorswith living cellular components,[ Proc. IEEE,vol. 91, no. 6, pp. 915–929, Jun. 2003.
[127] F. Heer, S. Hafizovic, W. Franks, A. Blau,C. Ziegler, and A. Hierlemann, BCMOSmicroelectrode array for bidirectionalinteraction with neuronal networks,[IEEE J. Solid-State Circuits, vol. 41, no. 7,pp. 1620–1629, Jul. 2006.
[128] S. Hafizovic, F. Heer, T. Ugniwenko,U. Frey, A. Blau, C. Ziegler, andA. Hierlemann, BA CMOS-basedmicroelectrode array for interaction withneuronal cultures,[ J. Neurosci. Methods,vol. 164, pp. 93–106, 2007.
[129] U. Frey, J. Sedivy, F. Heer, R. Pedron,M. Ballini, J. Mueller, D. Bakkum,S. Hafizovic, F. D. Faraci, F. Greve,K.-U. Kirstein, and A. Hierlemann,BSwitch-matrix-based high-densitymicroelectrode array in CMOS technology,[IEEE J. Solid-State Circuits, vol. 45, no. 2,pp. 467–482, Feb. 2010.
[130] L. Berdondini, P. D. van der Wal, O. Guenat,N. F. de Rooij, M. Koudelka-Hep, P. Seitz,R. Kaufmann, P. Metzler, N. Blanc, and
S. Rohr, BHigh-density electrode arrayfor imaging in vitro electrophysiologicalactivity,[ Biosens. Bioelectron., vol. 21,pp. 167–174, Jul. 15, 2005.
[131] R. Ehret, W. Baumann, M. Brischwein,M. Lehmann, T. Henning, I. Freund,S. Drechsler, U. Friedrich, M. L. Hubert,E. Motrescu, A. Kob, H. Palzer, H. Grothe,and B. Wolf, BMultiparametric microsensorchips for screening applications,[ FreseniusJ. Anal. Chem., vol. 369, pp. 30–35,Jan. 1, 2001.
[132] W. H. Baumann, M. Lehmann, A. Schwinde,R. Ehret, M. Brischwein, and B. Wolf,BMicroelectronic sensor system formicrophysiological application on livingcells,[ Sens. Actuators B, Chem., vol. 55,pp. 77–89, 1999.
[133] X. Y. Jiang, D. A. Bruzewicz, M. M. Thant,and G. M. Whitesides, BPalladium as asubstrate for self-assembled monolayers usedin biotechnology,[ Anal. Chem., vol. 76,pp. 6116–6121, Oct. 2004.
[134] R. Ehret, W. Baumann, M. Brischwein,A. Schwinde, and B. Wolf, BOn-line controlof cellular adhesion with impedancemeasurements using interdigitated electrodestructures,[ Med. Biol. Eng. Comput., vol. 36,pp. 365–370, May 1998.
[135] E. Thedinga, A. Kob, H. Holst, A. Keuer,S. Drechsler, R. Niendorf, W. Baumann,I. Freund, M. Lehmann, and R. Ehret,BOnline monitoring of cell metabolismfor studying pharmacodynamic effects,[Toxicol. Appl. Pharmacol., vol. 220, pp. 33–44,Apr. 2007.
[136] G. W. Gross, A. Harsch, B. K. Rhoades, andW. Gopel, BOdor, drug and toxin analysiswith neuronal networks in vitro:Extracellular array recording of networkresponses,[ Biosens. Bioelectron., vol. 12,pp. 373–393, 1997.
[137] G. W. Gross, B. Rhoades, and R. Jordan,BNeuronal networks for biochemicalsensing,[ Sens. Actuators B, Chem.,vol. B6, pp. 1–8, 1992.
[138] C. Ziegler, BCell-based biosensors,[ FreseniusJ. Anal. Chem., vol. 366, pp. 552–559,Mar./Apr. 2000.
[139] K. H. Gilchrist, V. N. Barker, L. E. Fletcher,B. D. DeBusschere, P. Ghanouni,L. Giovangrandi, and G. T. A. Kovacs,BGeneral purpose, field-portable cell-basedbiosensor platform,[ Biosens. Bioelectron.,vol. 16, pp. 557–564, 2001.
[140] Y. Papananos, T. Georgantas, and Y. Tsividis,BDesign considerations and implementationof very low frequencycontinuous-timeCMOS monolithic filters,[ Inst. Electr. Eng.Proc.VCircuits Devices Syst., vol. 144,pp. 68–74, 1997.
[141] R. R. Harrison and C. Charles, BA low-powerlow-noise {CMOS} amplifier for neuralrecording applications,[ Solid State Circuits,vol. 38, pp. 958–965, 2003.
[142] T. Delbruck and C. A. Mead, BAdaptivephotoreceptor with wide dynamic range,[ inProc. Int. Symp. Circuits Syst., 1994, vol. 4,pp. 339–342.
[143] J. J. F. Rijns, BCMOS low-distortionhigh-frequency variable-gain amplifier,[Solid State Circuits, vol. 31, pp. 1029–1034,1996.
[144] P. Livi, F. Heer, U. Frey, D. J. Bakkum, andA. Hierlemann, BCompact voltage andcurrent stimulation buffer for high-densitymicroelectrode arrays,[ in Proc. IEEE Int.Solid-State Circuits Conf., San Francisco, CA,2010, pp. 240–241.
Hierlemann et al.: Growing Cells Atop Microelectronic Chips: Interfacing Electrogenic Cells In Vitro
282 Proceedings of the IEEE | Vol. 99, No. 2, February 2011
[145] A. J. Lopez-Martin, S. Baswa,J. Ramirez-Angulo, and R.G. Carvajal,BLow-voltage super class AB CMOS OTAcells with very high slew rate and powerefficiency,[ IEEE J. Solid-State Circuits,vol. 40, no. 5, pp. 1068–1077, May 2005.
[146] A. Sedra and K. C. Smith, BAsecond-generation current conveyor and itsapplications,[ IEEE Trans. Circuit Theory,vol. CT-17, no. 1, pp. 132–134, Feb. 1970.
[147] F. Heer, W. Franks, A. Blau, S. Taschini,C. Ziegler, A. Hierlemann, and H. Baltes,BCMOS microelectrode array for themonitoring of electrogenic cells,[ Biosens.Bioelectron., vol. 20, pp. 358–366, Sep. 2004.
[148] R. F. Rice, ‘‘Some practical universalnoiseless coding techniques,[ JetPropulsion Lab., Pasadena, California,Mar. 1979, p. 79-22.
[149] J. Sedivy, U. Frey, S. Hafizovic, andA. Hierlemann, BMulti-chip high-densitymicroelectrode system for electrogenic-cellrecording and stimulation,[ in Proc. IEEESensors Conf., 2007, pp. 716–719.
[150] S. M. Potter, BDistributed processing incultured neuronal networks,[ Progr. BrainRes., vol. 130, pp. 49–62, 2001.
[151] S. I. Morefield, E. W. Keefer,K. D. Chapman, and G. W. Gross, BDrugevaluations using neuronal networkscultured on microelectrode arrays,[Biosens. Bioelectron., vol. 15, pp. 383–396,Oct. 2000.
[152] U. Frey, C. D. Sanchez-Bustamante,T. Ugniwenko, F. Heer, J. Sedivy,S. Hafizovic, B. Roscic, M. Fussenegger,A. Blau, U. Egert, and A. Hierlemann,BCell recordings with a CMOS high-densitymicroelectrode array,[ in Proc. Conf. IEEEEng. Med. Biol. Soc., 2007, vol. 2007,pp. 167–170.
[153] V. A. Alvarez and B. L. Sabatini,BAnatomical and physiological plasticityof dendritic spines,[ Annu. Rev. Neurosci.,vol. 30, pp. 79–97, 2007.
[154] S. M. Potter and T. B. DeMarse, BA newapproach to neural cell culture for long-termstudies,[ J. Neurosci. Methods, vol. 110,pp. 17–24, Sep. 30, 2001.
[155] G. J. Brewer, M. D. Boehler, T. T. Jones, andB. C. Wheeler, BNbActiv4 mediumimprovement to Neurobasal/B27 increasesneuron synapse densities and networkspike rates on multielectrode arrays,[ J.Neurosci. Methods, vol. 170, pp. 181–187,May 30, 2008.
[156] J. Hescheler, M. Halbach, U. Egert, Z. J. Lu,H. Bohlen, B. K. Fleischmann, andM. Reppel, BDetermination of electricalproperties of ES cell-derived cardiomyocytesusing MEAs,[ J. Electrocardiol., vol. 37,pp. 110–116, Oct. 2004.
[157] M. Reppel, P. Igelmund, U. Egert,F. Juchelka, J. Hescheler, and I. Drobinskaya,BEffect of cardioactive drugs on actionpotential generation and propagationin embryonic stem cell-derivedcardiomyocytes,[ Cell. Physiol. Biochem.,vol. 19, pp. 213–224, 2007.
[158] C. D. Sanchez-Bustamante, U. Frey,J. M. Kelm, A. Hierlemann, andM. Fussenegger, BModulation ofcardiomyocyte electrical propertiesusing regulated bone morphogeneticprotein-2 expression,[ Tissue Eng. A,vol. 14, pp. 1969–1988, 2008.
[159] U. Frey, U. Egert, F. Heer, S. Hafizovic, andA. Hierlemann, BSubcellular-resolutionrecording of electrical activity using aCMOS-microelectrode system,[ in Proc.IEEE Transducers, Denver, CO, 2009,pp. 1674–1677.
[160] R. H. Olsson, III, D. L. Buhl, M. N. Gulari,G. Buzsaki, and K. D. Wise, BA siliconmicroelectrode array for simultaneousrecording and stimulation in the brain ofbehaving animals,[ in Proc. 25th Annu. Int.Conf. IEEE Eng. Med. Biol. Soc., 2003, vol. 2,pp. 1968–1971.
[161] R. H. Olsson, III, D. L. Buhl, A. M. Sirota,G. Buzsaki, and K. D. Wise, BBand-tunableand multiplexed integrated circuits forsimultaneous recording and stimulation withmicroelectrode arrays,[ IEEE Trans. Biomed.Eng., vol. 52, no. 7, pp. 1303–1311, Jul. 2005.
[162] J. B. Bates and Y. T. Chu,BElectrode-electrolyte interface impedance:Experiments and model,[ Ann. Biomed. Eng.,vol. 20, pp. 349–362, 1992.
[163] X. Huang, D. Nguyen, D. W. Greve, andM. M. Domach, BSimulation ofmicroelectrode impedance changes due tocell growth,[ IEEE Sensors J., vol. 4, no. 5,pp. 576–583, Oct. 2004.
[164] E. A. Brown, J. D. Ross, R. A. Blum,Y. Nam, B. C. Wheeler, and S. P. DeWeerth,BStimulus-artifact elimination in amulti-electrode system,[ IEEE Trans. Biomed.Circuits Syst., vol. 2, no. 1, pp. 10–21,Mar. 2008.
[165] D. Bakkum, U. Frey, A. Hierlemann, andH. Takahashi, BSubcellular-resolutionelectronic recording and stimulation of
cultured cortical networks using an 11,011electrode CMOS array,[ Neurosci. Res.,vol. 65, p. S61-S61, 2009.
[166] D. Bakkum, U. Frey, J. Muller, M. Fiscella,H. Takahashi, and A. Hierlemann,BNovel neuronal cellular and networkmeasurements enabled by a high-density11’011-microelectrode CMOS array,[ in Proc.6th Int. Meeting Substrate-Integrated MicroElectrode Arrays, Reutlingen, Germany, 2010,pp. 324–326.
[167] FDA, ‘‘ICH S7B guideline: The nonclinicalevaluation of the potential for delayedventricular repolarization (QT intervalprolongation) by human pharmaceuticals,’’Rockville, MD, 20857-0001, 2004.
[168] T. Stieglitz and M. Gross, BFlexibleBIOMEMS with electrode arrangements onfront and back side as key component inneural prostheses and biohybrid systems,[Sens. Actuators B, Chem., vol. 83, pp. 8–14,Mar. 15, 2002, 2002.
[169] M. A. Nicolelis, BBrain-machine interfacesto restore motor function and probeneural circuits,[ Nat. Rev. Neurosci., vol. 4,pp. 417–422, May 2003.
[170] G. E. Perlin and K. D.-J. P. S. Wise,BUltra-compact integration forfully-implantable neural microsystems,[ inProc. IEEE 22nd Int. Conf. Micro Electro Mech.Syst., Sorrento, Italy, 2009, pp. 228–231.
[171] A. M. Sodagar, G. E. Perlin, Y. Yao,K. D. Wise, and K. Najafi, BAn implantablemicrosystem for wireless multi-channelcortical recording,[ in Proc. Transducers,2007, pp. 69–72.
[172] E. Zrenner, BWill retinal implants restorevision?’’ Science, vol. 295, pp. 1022–1025,Feb. 8, 2002.
[173] E. Zrenner, R. Wilke, K. U. Bartz-Schmidt,F. Gekeler, D. Besch, H. Benav,A. Bruckmann, K. Porubska, A. Kusnyerik,H. Sachs, T. Peters, B. Wilhelm,U. Greppmaier, A. Harscher, S. Kibbel,W. Wrobel, and A. Stett, BSubretinalmicroelectrode arrays allow blind retinitispigmentosa patients to recognize lettersand combine them to words,[ in Proc.2nd Int. Conf. Biomed. Eng. Inf., 2009,DOI: 10.1109/BMEI.2009.5305315.
[174] G. E. Loeb, BCochlear prosthetics,[ Annu.Rev. Neurosci., vol. 13, pp. 357–371, 1990.
[175] J. P. Rauschecker and R. V. Shannon,BSending sound to the brain,[ Science,vol. 295, pp. 1025–1029, Feb. 8, 2002.
ABOUT T HE AUTHO RS
Andreas Hierlemann (Member, IEEE) received
the Diploma in chemistry from the University of
Tubingen, Tubingen, Germany, in 1992 and the
Ph.D. degree in physical chemistry from the
Eberhard-Karls University, Tubingen, Germany,
in 1996.
After that, he held two postdoctoral positions
at the Texas A&M University, College Station
(1997) and at the Sandia National Laboratories,
Albuquerque, NM (1997–1998). In 1999, he joined
the Department of Physics, ETH Zurich, Zurich, Switzerland, where he
was appointed Associate Professor of Microsensorics in June 2004. In
April 2008, he became Full Professor of Biosystems Engineering at the
Department of Biosystems, Science and Engineering (D-BSSE) of ETH
Zurich in Basel, Switzerland. His research interests include the develop-
ment of integrated chemical and biomicrosensor systems, the develop-
ment of microfluidic techniques for cell handling and cell characterization,
and the direct coupling of biological entities, such as neurons or heart cells,
to microelectronic chips.
Urs Frey received the diploma in electrical
engineering from ETH Zurich, Basel, Switzerland,
in 2003 and the Ph.D. degree in electrical engi-
neering from the Physical Electronics Laboratory,
the Department of Information Technology and
Electrical Engineering, ETH Zurich, in 2008.
From 2008 to 2009, he was with the Bio
Engineering Laboratory, ETH Zurich, where he
was leading the CMOS-based MEA activities. In
2009, he joined IBM Research, Zurich, Switzerland.
The focus of his activities is on system design and interface circuitry for
CMOS-based sensors and storage devices.
Hierlemann et al. : Growing Cells Atop Microelectronic Chips: Interfacing Electrogenic Cells In Vitro
Vol. 99, No. 2, February 2011 | Proceedings of the IEEE 283
Sadik Hafizovic received the diploma in micro-
system technology from IMTEK, University of
Freiburg, Freiburg, Germany, in 2002.
He worked on his diploma thesis at the Tabata
Laboratories, Ritsumeikan University, Japan. From
2002 to 2007, he was with the Physical Electronics
Laboratory, ETH Zurich, Basel, Switzerland. From
2008 to 2010, he was a Researcher at the Bio
Engineering Laboratory, ETH Zurich. The focus of
his research activities was on atomic force
microscopy and CMOS-based microelectrode arrays for neuronal inter-
faces. Currently, he is CEO of Zurich Instruments, Zurich, Switzerland.
Flavio Heer (Member, IEEE) received the diploma
in physics and the Ph.D. degree from ETH Zurich,
Basel, Switzerland, in 2001 and 2005, respectively.
Since 2005, he has been a Senior Researcher at
the Bio Engineering Laboratory, ETH Zurich,
where he is involved in system design and
interface circuitry for CMOS-based biosensors
and biochemical sensors. His research interests
include signal processing and circuit design for
sensor interfaces. Moreover, he is a cofounder of
Zurich Instruments, Zurich, Switzerland.
Hierlemann et al.: Growing Cells Atop Microelectronic Chips: Interfacing Electrogenic Cells In Vitro
284 Proceedings of the IEEE | Vol. 99, No. 2, February 2011