Transparent, conformable, active multielectrode arrayusing organic electrochemical transistorsWonryung Leea, Dongmin Kima,b, Naoji Matsuhisaa, Masae Nagasea,b, Masaki Sekinoa,b, George G. Malliarasc,Tomoyuki Yokotaa,b, and Takao Someyaa,b,d,e,1

aDepartment of Electrical Engineering and Information Systems, The University of Tokyo, Tokyo 113-8656, Japan; bExploratory Research for AdvancedTechnology, Japan Science and Technology Agency, Tokyo 113-8656, Japan; cDepartment of Bioelectronics, Ecole Nationale Supérieure des Mines, 13541Gardanne, France; dThin-Film Device Laboratory, RIKEN, Saitama 351-0198, Japan; and eCenter for Emergent Matter Science, RIKEN, Saitama 351-0198, Japan

Edited by John A. Rogers, Northwestern University, Evanston, IL, and approved August 22, 2017 (received for review March 8, 2017)

Mechanically flexible active multielectrode arrays (MEA) have beendeveloped for local signal amplification and high spatial resolu-tion. However, their opaqueness limited optical observation andlight stimulation during use. Here, we show a transparent, ultra-flexible, and active MEA, which consists of transparent organicelectrochemical transistors (OECTs) and transparent Au grid wir-ings. The transparent OECT is made of Au grid electrodes and hasshown comparable performance with OECTs with nontransparentelectrodes/wirings. The transparent active MEA realizes the spatialmapping of electrocorticogram electrical signals from an optoge-netic rat with 1-mm spacing and shows lower light artifacts thannoise level. Our active MEA would open up the possibility ofprecise investigation of a neural network system with directlight stimulation.

organic electrochemical transistors | thin films | multielectrode array |optogenetic | transparent electrode

Amultielectrode array (MEA), which is a device consisting ofa 2D array of microelectrodes, is widely used to detect the

action potential of neurons and/or muscle cells (1–3). Thecharacterization of neurons with MEA enables one to accuratelymeasure positions of active/inactive cells, propagation of neuralsignals, and/or networking among multiple neurons, which can-not be measured by single-point detection (4, 5). The applica-tions of MEA can be classified into two groups—namely, in vitroarrays manufactured on glass and/or petri dishes (6, 7) andin vivo arrays made of microwires (8) and/or silicon-based nee-dles (9). The state of the art microneedle array for detecting theaction potential of the brain has realized high spatial resolutionof 200 μm (10) and high temporal resolution of 60 μs (11) si-multaneously. Recently, a mechanically flexible MEA has beendeveloped to realize conformal contact between curved surfacesof the brain and electrodes with minimal invasiveness (3, 8, 12,13). The flexible MEA has realized simultaneous recording oflocal field potentials and action potentials from the humancortical surface without penetrating the brain tissues (14).To advance MEA, an active electrode array, in which each

passive electrode is coupled with an amplifier, has been exam-ined to improve signal integrity (15–17). Recently, the organicelectrochemical transistor (OECT) has been intensively studiedas an active component for the purposes of detection and am-plification of biosignals. The mixed electronic and ionic transportin the active conducting polymer layer enables extremely hightransconductance of ∼1 mS as well as fast response speed of∼1 kHz (18, 19). Because of the aforementioned properties,EEG (18), electrocorticography (ECoG) (17), and ECG (20)have all been measured using OECTs, with high signal-to-noiseratio (SNR) of at least 54 dB (17). Furthermore, integration ofOECTs acting as sensors and organic field effect transistors(OFETs) as a multiplexer was used to form an active electrodearray (21). In addition to a function of signal amplification, anactive matrix design can significantly reduce the total number ofwirings to access each microelectrode, realizing a scalable MEA

design. Such an active matrix addresser, realized on a flexiblesubstrate, has been well-researched for various active compo-nents, such as OFETs, silicon membrane transistors, and nano-wire transistors (2, 22–28).However, optical transparency is needed in MEA for observ-

ing biotissues using optical microscopy or modulation of ionsimaging (29, 30). For this purpose, thin metal electrodes withtypical thickness of ∼3–7 nm are used to achieve transparencywith transmittance of ∼30–70% (31). Because thin metal cannotachieve high transmittance and high conductance simulta-neously, fully transparent conductors are needed. From thisviewpoint, indium-tin-oxide (ITO) (32) and graphene (30, 33)were used to realize a highly transparent MEA comprisingtransparent electrodes and transparent wires. Such a hightransparency is needed particularly for MEA with the largenumber of multiple electrodes, because wirings occupy mostof the area. Furthermore, transparent MEA will create a newtool that can stimulate cells and detect signals via both lightand electricity.However, a transparent and active MEA has not yet been re-

alized owing to two technical difficulties. First, many transparentconductors and transistors exhibit large changes of electric per-formances under the illumination of light. Such photosensitivity,which is usually ascribed to the charge traps or band gaps intransparent materials (34, 35), makes it difficult to detect smallchanges of active potentials of neurons by photoexcitation. Sec-ond, the integration of transparent conductors and transparenttransistors requires sophisticated engineering of device designs,because transparency and conductivity of transparent conductorshave a tradeoff relation (36, 37). The optimized device structuremust be designed such that process compatibility may be main-tained. This requirement is much more difficult to meet whenMEA is manufactured on flexible substrate owing to its lowprocess temperature.

Significance

We have developed a transparent, ultraflexible, and activemultielectrode array (MEA), which consists of transparent or-ganic electrochemical transistors and transparent Au grid wir-ings. The micropatterned Au grid showed 60% transparency at475-nm wavelength. The transparent active MEA showed thespatial mapping of electrocorticogram electrical signals from anoptogenetic rat with 1-mm spacing and shows lower light ar-tifacts than noise level.

Author contributions: W.L., N.M., G.G.M., T.Y., and T.S. designed research; W.L., D.K.,M.N., and M.S. performed research; W.L. contributed new reagents/analytic tools; W.L.analyzed data; and W.L. and T.S. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.1To whom correspondence should be addressed. Email: someya@ee.t.u-tokyo.ac.jp.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1703886114/-/DCSupplemental.

10554–10559 | PNAS | October 3, 2017 | vol. 114 | no. 40 www.pnas.org/cgi/doi/10.1073/pnas.1703886114

Dow

nloa

ded

by g

uest

on

Dec

embe

r 8,

202

0

In this work, we have successfully manufactured a transparentand active MEA on a 1.2-μm-thick parylene substrate. The de-vice consists of transparent OECTs and transparent metal gridwirings, realizing high transparency on the entire device area.The Au grid wiring with linewidth of 3 μm, which is much smallerthan the typical size of neurons, exhibits high transparency of60% and low sheet resistance of 3 Ω/sq on 1.2-μm-thick parylenesubstrate. The mechanical durability of the Au grid was per-formed by applying compression strain on 100% prestretchedelastomer. The current fluctuation of OECT is suppressed to lessthan 0.1% when a strong laser beam (150 mW) is illuminated onthe channel of transparent OECTs. Furthermore, a 3 × 5 array ofOECT has been manufactured with a total thickness of 3 μm,where each OECT exhibits large transconductance (gm) of1.1 mS and fast response time of 363 μs. Finally, the feasibilityof a transparent and active MEA has been shown by mappingevoked response having an amplitude of ≈800 μV at the exactposition where the cortical surface of an optogenetic rat isstimulated by a laser beam.

ResultsThe transparent MEA consisting of OECTs and Au metal gridwirings is manufactured on a 1.2-μm-thick parylene substrate(Fig. 1A). The total thickness is as small as 3 μm, including thetop encapsulation of a 1.2-μm-thick photoresist epoxy (SU-8)layer, resulting in high flexibility and conformability (8, 38).The fabrication process is shown in Methods and SI Appendix,Fig. S1 in detail. The active layer of OECT is inherently trans-parent owing to the intrinsic transparency of thin poly(2,3-dihy-drothieno-1,4-dioxin)-poly(styrenesulfonate) (PEDOT:PSS) activelayers, whereas the transparency of wirings and electrodes is re-alized by Au metal grids (Fig. 1B). Fig. 1C shows the cross-sectionof a transparent OECT. Fig. 1D shows the circuit diagram thatenables selective access to each OECT by applying voltage to theselected line. A photograph of the 3 × 5 OECT array that is con-formal onto a rat brain shows that the cerebrovascular of the braincan be recognized clearly through the devices (Fig. 1E).The Au grid electrodes are carefully characterized to ensure

optimal transparency and conductivity. Fig. 2A shows the opticalmicroscopic image of transparent and flexible Au grid elec-trodes. The transparency and sheet resistance depend on thewidth (w) of the grid and periodicity (p) of the unit (Fig. 2B). Theoptical and electrical characteristics are investigated by changingp among 12, 15, and 18 μm, while w is fixed at 3 μm, which is theminimal resolution of our lithography system. Fig. 2C shows themeasured transparency (Tm) of the Au grids that are laminatedon the supporting glass substrate. The opening aperture ratio(Tc) is defined as (p − w)2/p2. After subtracting the transparencyof the glass substrate, Tm is linearly proportional to Tc. Becausetransparency and conductivity of the metal grid have a tradeoffrelationship (SI Appendix, Fig. S2), w and p must be optimallydesigned. Indeed, low sheet resistance of wires is important tominimize the voltage drop in the wire and subsequently, to applysufficient voltage from the contact pads to the wired OECTchannel. In the following experiment, w and p are set to be 3 and18 μm, respectively. In this design, Tc is 69%, which is comparedwith that of the ITO MEA for optogenetic applications (32),whereas the sheet resistance is 3 Ω/sq, which is low enough tooperate OECTs.In addition to the high transparency and high conductivity

mentioned above, another attribute of Au grids is good me-chanical durability. To characterize mechanical durability, Augrids with the thickness of 70 nm are formed on 1.2-μm parylenesubstrates and covered by a 150-nm-thick PEDOT:PSS layer. Forcomparison, the reference sample, replacing 70-nm-thick Augrids with 70-nm-thick plane ITO, is also prepared. These de-vices are laminated onto elastomer that is prestretched by 100%.After the lamination, the compression strain of prestretched

elastomer is slowly released to form wrinkling structures on thesurface of the elastomer. Fig. 2D shows the microscopic imagesof Au grids on elastomer without compression and with com-pression strain of 30%. The Au grid and ITO with PEDOT:PSSwere patterned with dimensions of 1 × 1 cm. Furthermore, aplain 100-nm-thick Au was deposited at the edge of the sheet ofdimensions of 1 × 1 cm, with a contact pad of 1 × 0.5 cm. Thesheet resistance of the 1 × 1-cm Au grid and ITO was measuredunder dry conditions using the two probes method. The sheetresistance is measured with increasing compression strain from0 to 50%. The sheet resistance of the Au grid sample changesfrom 3 to 7 Ω/sq at the compression strain of 50%, while that ofthe ITO sample increases from 79 to 378 Ω/sq (Fig. 2E). Thecyclic test is then carried out, applying compression strain of 50%(Fig. 2F). The resistance of the Au grid sample increases by 40Ω/sq,whereas that of the ITO sample increases by 8 kΩ/sq, showingexcellent mechanical durability of the Au grids.The transparent OECTs made with the Au grids are prepared,

and their I–V characteristics and time response are examined.For comparison, OECTs with nontransparent Au plane filmswithout mesh structures, which are referred to as nontransparent

ElectrolyteOECT

Data lines

Scan lines

device

Data lines

Scanlines

Interconnection

Au grid

PEDOT:PSS

Su-8 (passivation)

SubstrateParylene (1200 nm)

Source/DrainAu grid (70 nm)

PEDOT:PSS (140 nm)

1st passivation for interconnectionParylene (600 nm)

2nd passivationSu-8 (1200 nm)

A

C

D E

B

Fig. 1. Transparent electrophysiology OECTs array. (A) Image of the trans-parent electrophysiology OECT array on a parylene substrate. The passiv-ation layer was patterned by SU-8. (B) Magnified image of a transparentOECT single cell. For transparency, the source and drain of OECT were madeby a metal grid of Au. (C) A cross-section of the transparent OECT array.(D) Schematic diagram of the OECT array. A certain voltage was applied tothe drain line to be measured. The other lines were all connected to theground of the circuit for preventing cross-talk. The source lines wereconnected to each current meter. (E ) The 3 × 5 transparent electrophysi-ology array (black dashed square) on the cortical surface of the opto-genetic rat. (Scale bar: 1 mm.)

Lee et al. PNAS | October 3, 2017 | vol. 114 | no. 40 | 10555

APP

LIED

PHYS

ICAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Dec

embe

r 8,

202

0

OECTs, are also prepared. The channel width and length are setto be 90 and 60 μm, respectively (SI Appendix, Fig. S3). Thefabrication process is detailed in Methods and SI Appendix, Fig.S1. The measurement setup is schematically shown in Fig. 3A,where a gate bias of OECTs is applied via PBS solution and anAg/AgCl electrode. Fig. 3B shows the transfer curves. The gmvalues of transparent and nontransparent OECTs (gm = ΔId/ΔVg)are as high as 2.3 and 2.2 mS, respectively, at gate voltage of 0 V.Note that the difference in gm is only 4%. The response times (τ)of the transparent and nontransparent OECTs are measured byapplying a gate voltage pulse with a duration of 1 ms (Fig. 3C). τis evaluated by fitting experimental data with an exponentialfunction. The τ values of transparent and nontransparent OECTswere as low as 97 and 112 μs, respectively, with a gate voltagepulse of 100 mV (Fig. 3B). The difference in τ is only 13%. Thecutoff frequency for this device was 50 Hz, with the trans-conductance of more than 1 mS at 100 Hz (SI Appendix, Fig. S4).This is sufficient to measure the low-frequency ECoG of anoptogenetic animal induced by light (39). Moreover, the leakagecurrent of SU-8 was measured with dc bias of 0.6 V (SI Appendix,Fig. S5). The leakage current of SU-8 was low at 8 nA/cm2 aftersoaking in PBS of 37 °C with the bias of 0.6 V for 24 h, although itshowed poor sealing property when used as the passivation layerfor reactive materials, such as Mg (40). Also, the gate leakagecurrent was low at 50 nA after soaking in PBS of 37 °C for 24 h (SI

Appendix, Fig. S6). Therefore, the SU-8 passivation is sufficient forshort-term surgical applications.A 3 × 5 array of transparent OECTs is fabricated with 1-mm

spacing (Fig. 3D). The channel width and length are 70 and20 μm, respectively (Fig. 3D, Inset). Fig. 3 E and F shows thestatistics of the transconductance and the response time for thearray. The average transconductance and response time are1.1 mS and 340 μs, respectively. The transconductance of thearray is slightly lower than that of the discrete devices because ofthe channel dimension and resistance of the interconnection.With the lengths of the data and scan lines of 3 and 5 mm, re-spectively, and their width of 0.5 mm, the highest resistance ofthe grid interconnection can be estimated to be ≈48 Ω. In-creasing the thickness of the gold layer beyond the 70 nm wouldallow for a larger and denser array of transparent OECTs (SIAppendix, Fig. S2).In our array design with the simplest structure in which one

cell is formed by one OECT, the cross-talk among sensor cellscan be avoided by applying appropriate voltage biases (SI Ap-pendix, Figs. S7 and S8), which is confirmed by the circuit sim-ulator (SI Appendix, Figs. S9 and S10) and an experiment (SIAppendix, Fig. S11) for a 3 × 3 OECT array. The switching of ourcircuit can be controlled not by the gate bias (Vgs) but by thedrain bias (Vds). In addition, the load resistance (RL) of thewaveform generator/fast measurement unit (WGFMU), whichwas used for recording of the current, and the wire resistance

100

101

102

103

0 10 20 30 40 50

ITO + PEDOT:PSSAu grid + PEDOT:PSS

( ecnatsiseR

Ω)qs /

Compression (%)

020406080

100120140

0 20 40 60 80 100

Au grid + PEDOT:PSSITO + PEDOT:PSS

R/R

0

Cycle number(Compression strain 50%)

0

20

40

60

80

100

300 400 500 600 700 800

)%( ycneraps nar T

Wavelength λ (nm)

w

p

Tc = 0,44,62,69

glass

Compression 30%

B C

E F

A

D

Compression 0%

Fig. 2. Mechanical stability and transparency of Au grid. (A) Microscope image of metal grid on 1.2-μm parylene substrate. (Scale bar: 100 μm.) (B) Image of ametal grid and square grid unit cell parameters. (C) Transmittance spectrum of the metal grid with various calculated transmittance (Tc). (D) A 3D microscopeimage of a metal grid under (Upper) 0% compression strain and (Lower) 30% compression strain. (E) Resistance change of ITO (70 nm) and Au grid (70 nm)with PEDOT:PSS (150 nm) when compression strain was changed (the number of samples was three). (F) Resistance ratio after cycling test of ITO and Au gridwith PEDOT:PSS when the compression strain applied to the film was 50%.

10556 | www.pnas.org/cgi/doi/10.1073/pnas.1703886114 Lee et al.

Dow

nloa

ded

by g

uest

on

Dec

embe

r 8,

202

0

(RW) of the Au grid were low at ≈0 and 10 Ω, respectively. Thelow RL and RW minimize the cross-talk, while the cross-talk wasnoticeable when high RL and RW were introduced (SI Appendix,Figs. S12 and S13). The data line in the array is chosen by ap-plying drain voltage (−0.6 V) to the selected line, while the otherlines are connected to ground (0 V). The cross-talk evaluation isperformed by measuring a cell or a data line while a sine wave ofdifferent frequencies is applied to each gate voltage and a drainvoltage (−0.6 V) is applied to the measured cell. Both the circuitsimulation (SI Appendix, Figs. S8–S10) and the experiment (SIAppendix, Fig. S11) showed clear current fluctuation of the sinewave, which has the same frequency as the measured cell.The feasibility of the transparent OECT array has been shown

by ECoG recordings on the cortical surface of optogenetic rats(41, 42). The rats express a light-sensitive gene, channelrodopsin-2,in their nerves and their terminals (43). The ECoG signals areinduced by a laser beam at the wavelength of 473 nm. To inducelocalized brain activity, the rat’s brain surface is surgically exposedand targeted by fiber-guided laser stimulation. The 3 × 5 array oftransparent OECTs is attached to the exposed surface of the brainand individually measured signals evoked by light stimulationthrough the device. The conformal contact between the devicefilm and the surface of the brain is achieved by the ultraflexibilityof thin films.The laser beam illuminates the exact spot on the channel re-

gion of OECTs. Fig. 4A shows the photograph of the transparentOECT laminated on the cortical surface of the rat with the laser

stimulation (40 mW). Owing to high transparency, the reflectionof the laser stimulation on the device surface is minimized. Fig. 4B and C shows the ΔIds/gm under laser stimulations with differentexcitation intensities of 17, 40, 60, and 110 mW for transparentand nontransparent OECTs, respectively. The result for thetransparent OECT exhibits double the amplitude (700 μV) com-pared with the nontransparent OECT (350 μV) at the same lightintensity of 40 mW. The rms values were 38 and 60 μV fortransparent OECT and nontransparent OECT, respectively.With increasing the laser power from 40 to 110 mW, the dif-ference in electric signals between transparent and nontransparentOECTs decreases from 200 to 150% because of the saturation ofthe photoresponsivity at high intensity.To exclude the possibilities of light artifacts, such as the

Becquerel effect or light response of active materials, we mea-sured the evoked potentials, with the rms of 0.03 μA, in responseto a pair of light pulses (40 mW) with intervals of 4, 8, 12, and20 ms (Fig. 4D). Because the repolarization of the first neuralexcitation does not occur within 4 ms, one signal should bemeasured by a pair of pulses with an interval of 4 ms or less. Onepeak of the signal is recorded at intervals of 4 ms. With in-creasing interval, splitting of signals is clearly observed, becausethe repolarization of the first neural excitation has been com-pleted. These electrical responses associated with repolarizationtime, which are known as a characteristic of neurons (39), un-ambiguously prove that the signals are not light artifacts butactual neural activations. It is important to note that the hindrance

A

D E F

B C

Fig. 3. Electrical characteristic of transparent OECT and OECT array. (A) OECT during electrical measurement. The electrolyte was placed on the channel ofOECTs, and Vg was applied using Ag/AgCl from the top of the electrolyte. (B) Transfer characteristics of the transparent and nontransparent OECTs. Thechannel width and length were 90 and 20 μm, respectively. The maximum transconductance values were 2.2 and 2.3 mS at a −0.7-V source drain and a 0-Vgate voltage, respectively. (C) The response time of transparent and nontransparent OECT. The response times were 110 and 120 μs, respectively. (D) Themicroscope image of a 3 × 5 transparent OECT array and magnified view of the channel (Inset). The channel width and length were 70 and 20 μm, respectively.(Scale bar: D, 500 μm; D, Inset, 30 μm.) (E) The distribution of transconductance in a 3 × 5 transparent OECT array. (F) The distribution of response times in a 3 ×5 transparent OECT array.

Lee et al. PNAS | October 3, 2017 | vol. 114 | no. 40 | 10557

APP

LIED

PHYS

ICAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Dec

embe

r 8,

202

0

of the artifact by light stimulation is comparable with the peak-to-peak noise level of 0.1 μA (SI Appendix, Fig. S14). Nonethe-less, the light is targeted directly onto devices. Because lightartifacts are a common issue in optogenetic recordings (30),additional discussion can be found later in the text.Using a 3 × 5 array of transparent OECTs with 1-mm2 spatial

resolution, mapping of activation potentials is measured. Thedevice is placed on the cortical surface of an optogenetic rat. Thelaser beam from the optical fiber stimulates the center spot ofthe array (Fig. 4E). Fig. 4F indicates a strong ECoG signal withan amplitude of 0.96 μA measured by the OECT at the center ofthe array. The SNR was 13 dB, whereas the peak-to-peak noisewas 0.2 μA, and the rms noise level was recorded as 0.02 μA.Signals are also obtained at the upper and lower cells neigh-boring the central cell owing to the diffusion of laser fibers.

DiscussionHere, we discuss the scalability of our circuit design. Previously, a4 × 4 array of OECT had been fabricated, in which each of theOECTs was individually wired (44). Because an N × N array in theindividually wired design requires N2 wires, the area is dominatedby wirings for a large N. In contrast, our active matrix design canreduce the number of wirings to 2N, realizing scalability for a largeN. Note that our array has a simple design without access tran-sistors. However, it does not suffer from cross-talk issues, because

the current fluctuation of OECTs is measured without load im-pedance. Another type of active matrix design—each cell con-sisting of one OECT and one access transistor—can reduce thepower consumption and the cross-talk induced by RL and RW aswell as simplify the readout circuit (21, 25, 26, 45); thus, it isexpected that integration of transparent OECT and access tran-sistor will further improve the scalability from the viewpoint ofpower consumption and improved resolution of measurement.Finally, we would like to emphasize again that the key to re-

alize transparent active MEA is the elimination of light artifacts,which are frequently major problems in optogenetic neural signalrecordings. Note that photosensitivity is suppressed in OECT fortwo reasons. First, the absorption of PEDOT:PSS in OECTs issignificantly reduced in the blue-wavelength regime (46), be-cause highly doped holes in PEDOT:PSS compensate for the siteof the charge trap. Second, the Becquerel effect, which is in-versely proportional to the site impedance, is suppressed by thelow site impedance of PEDOT:PSS (47). Indeed, it is reportedthat a Pt electrode with site impedance of 50 kΩ at 1 kHz couldreduce the light artifact (39). Our PEDOT:PSS layer exhibits alower site impedance of 10 kΩ at 1 kHz and 1,400 μm2 (SI Ap-pendix, Fig. S15) owing to high conductivity and the porousstructure (48). In this way, our device showed no detectable lightartifacts, even under high-power laser stimulation (110 mW) (SIAppendix, Fig. S14).

A B

D E F

C

Fig. 4. In vivo evaluation of transparent OECTs with light stimulation using optogenetic rat. (A) Photograph of the transparent OECT on a neuron-con-centrated area of the cortical surface of optogenetic mice with blue laser continuous stimulation through optical fiber (500-μm diameter). (Scale bar: 500 μm.)(B) The recorded signal by transparent OECT with light stimulation of a 475-nm wavelength. The rms was 38 μV. (C) The recorded signal by nontransparentOECT with light stimulation of a 475-nm wavelength. The rms was 60 μV. (D) Evoked ECoG potentials in response to pairs of 2-ms light pulses at 60 mW. Therms was 0.03 μA. (E) Photograph of the transparent electrophysiology array in a neuron-concentrated area on the cortical surface of optogenetic mice withblue laser continuous stimulation through optical fiber (500-μm diameter). (Scale bar: 1 mm.) (F) The spatial distribution of the brain signal intensity measuredby a 3 × 5 transparent electrophysiology array. The electric potential was calculated using each OECT’s transconductance, which is measured by a Vg sinusoidalsignal and Id.

10558 | www.pnas.org/cgi/doi/10.1073/pnas.1703886114 Lee et al.

Dow

nloa

ded

by g

uest

on

Dec

embe

r 8,

202

0

MethodsDevice Fabrications. Novec 1700 (3M) dissolved in Novec 7100 (3M) was spin-coated on a glass wafer as a sacrificial layer. Parylene dix-SR was deposited asan ultrathin substrate of the device to fabricate 1.2-μm-thick films; 70-nm-thick Au formed the pattern via the liftoff process. The device wasencapsulated by 600 nm of parylene to separate the layer for in-terconnection; 70-nm-thick Au for interconnection was deposited by theliftoff process. For passivation, the 1.2-μm-thick SU-8(3005) was patternedby photolithography. The solution of PEDOT:PSS with additives (17) wasspin-coated on the film at 2,000 rpm (Spincoater 1H-DX2, MIKASA). Finally,the PEDOT:PSS was patterned by dry-etching (49).

Characterization. All characterization of the OECTs was performed using asolution of PBS as the electrolyte and an Ag/AgCl wire (Warner Instruments)as the gate of the OECT electrode. The I–V characteristics of the OECTsand the OFETs were measured with a 4155C (Agilent). The WGFMUmodule of the B1500 (Keysight) measured the drain currents; the otherWGFMU module measured the gate voltages of the OECT to determinethe time response. A multifunction generator was used to generate thesinusoidal waves.

In Vivo Evaluation. Animal experiments were conducted in accordance withthe Guidelines of the Animal Experiment Committee at the University ofTokyo. The rat [W-Tg(Thy1-COP4-YFP*)4Jfhy; National BioResource Project ofthe Rat in Japan; male, 12 wk old, 268 g] was anesthetized using 2–2.5%

isoflurane mixed with air, and the brain skull was incised to expose a corticalsurface. Stimulation light with a wavelength of 473 nm was delivered from alaser source (COME2-LB473/586/200S; LUCIR). The neural signal was mea-sured by connecting a semiconducting parameter analyzer (the WGFMUmodule of the B1500; Keysight) to the attached OECT device while opticallystimulating the nerve terminal at repetitive rates of 2 Hz. The neural re-cording from matrix OECT was measured individually with the sampling rateof 1 kHz. The measured data were obtained from a single trace withoutaveraging. The filtration of the noise above 1 kHz was obtained, imple-mented from the following formula:

yn =�

1f−3dB

+ ts

�yn−1f−3dB

+�

1f−3dB

+ ts

� ðxn − xn−1Þf−3dB

, [1]

where x, ts, and f−3dB denote the raw data, the interval time, and cutofffrequency, respectively.

ACKNOWLEDGMENTS. We thank Marce Ferro, Pierre Leleux, JonathanRivnay, Junhyung Kim, Sunghoon Lee, Robert Nawrocki, Sungjoon Park,and Prof. Makoto Takamiya for the discussions and Prof. Hiromu Yawo forsupplying the optogenetic rat. W.L. was supported by the Japan Society forthe Promotion of Science (JSPS) through the Program for Leading GraduateSchools. N.M. was supported by the Advanced Leading Graduate Course forPhoton Science and a JSPS research fellowship for young scientists. This workwas supported by Japan Science and Technology Agency ExploratoryResearch for Advanced Technology Grant JPMJER1105.

1. Rubehn B, Bosman C, Oostenveld R, Fries P, Stieglitz T (2009) A MEMS-based flexiblemultichannel ECoG-electrode array. J Neural Eng 6:036003.

2. Viventi J, et al. (2011) Flexible, foldable, actively multiplexed, high-density electrodearray for mapping brain activity in vivo. Nat Neurosci 14:1599–1605.

3. Minev IR, et al. (2015) Electronic dura mater for long-term multimodal neural inter-faces. Science 347:159–163.

4. Sawada M, et al. (2015) Function of the nucleus accumbens in motor control duringrecovery after spinal cord injury. Science 350:98–101.

5. Ishida A, et al. (2016) Causal link between the cortico-rubral pathway and functionalrecovery through forced impaired limb use in rats with stroke. J Neurosci 36:455–467.

6. Oka H, Shimono K, Ogawa R, Sugihara H, Taketani M (1999) A new planar multi-electrode array for extracellular recording: Application to hippocampal acute slice.J Neurosci Methods 93:61–67.

7. Heuschkel MO, Fejtl M, Raggenbass M, Bertrand D, Renaud P (2002) A three-dimensional multi-electrode array for multi-site stimulation and recording in acutebrain slices. J Neurosci Methods 114:135–148.

8. Kim D-H, et al. (2010) Dissolvable films of silk fibroin for ultrathin conformal bio-integrated electronics. Nat Mater 9:511–517.

9. Buzsáki G (2004) Large-scale recording of neuronal ensembles. Nat Neurosci 7:446–451.10. Wark HAC, et al. (2013) A new high-density (25 electrodes/mm2) penetrating mi-

croelectrode array for recording and stimulating sub-millimeter neuroanatomicalstructures. J Neural Eng 10:45003.

11. Harrison RR, et al. (2007) A low-power integrated circuit for a wireless 100-electrodeneural recording system. IEEE J Solid-State Circuits 42:123–133.

12. Kim D-H, et al. (2011) Epidermal electronics. Science 333:838–843.13. Liu Y, et al. (2016) Epidermal mechano-acoustic sensing electronics for cardiovascular

diagnostics and human-machine interfaces. Sci Adv 2:e1601185.14. Khodagholy D, et al. (2015) NeuroGrid: Recording action potentials from the surface

of the brain. Nat Neurosci 18:310–315.15. Fuketa H, et al. (2014) 1 μm-thickness ultra-flexible and high electrode-density surface

electromyogram measurement sheet with 2 V organic transistors for prosthetic handcontrol. IEEE Trans Biomed Circuits Syst 8:824–833.

16. Lacour SP, Jones J, Wagner S, Li T, Suo Z (2005) Stretchable interconnects for elasticelectronic surfaces. Proc IEEE 93:1459–1467.

17. Khodagholy D, et al. (2013) In vivo recordings of brain activity using organic tran-sistors. Nat Commun 4:1575.

18. Rivnay J, et al. (2015) High-performance transistors for bioelectronics through tuningof channel thickness. Sci Adv 1:e1400251.

19. Khodagholy D, et al. (2013) High transconductance organic electrochemical transis-tors. Nat Commun 4:2133.

20. Campana A, Cramer T, Simon DT, Berggren M, Biscarini F (2014) Electrocardiographicrecording with conformable organic electrochemical transistor fabricated on re-sorbable bioscaffold. Adv Mater 26:3874–3878.

21. Lee W, et al. (2016) Integration of organic electrochemical and field-effect transistors forultraflexible, high temporal resolution electrophysiology arrays. Adv Mater 28:9722–9728.

22. Kim J, et al. (2016) A wearable multiplexed silicon nonvolatile memory array usingnanocrystal charge confinement. Sci Adv 2:e1501101.

23. Fang H, et al. (2017) Capacitively coupled arrays of multiplexed flexible silicon tran-sistors for long-term cardiac electrophysiology. Nat Biomed Eng 1:38.

24. Yu KJ, et al. (2016) Bioresorbable silicon electronics for transient spatiotemporalmapping of electrical activity from the cerebral cortex. Nat Mater 15:782–791.

25. Someya T, et al. (2004) A large-area, flexible pressure sensor matrix with organic field-effect transistors for artificial skin applications. Proc Natl Acad Sci USA 101:9966–9970.

26. Kawahara J, et al. (2013) Flexible active matrix addressed displays manufactured byprinting and coating techniques. J Polym Sci B Polym Phys 51:265–271.

27. Takei K, et al. (2010) Nanowire active-matrix circuitry for low-voltage macroscaleartificial skin. Nat Mater 9:821–826.

28. Schwartz G, et al. (2013) Flexible polymer transistors with high pressure sensitivity forapplication in electronic skin and health monitoring. Nat Commun 4:1859.

29. Williamson A, et al. (2015) Localized neuron stimulation with organic electrochemicaltransistors on delaminating depth probes. Adv Mater 27:4405–4410.

30. Park DW, et al. (2014) Graphene-based carbon-layered electrode array technology forneural imaging and optogenetic applications. Nat Commun 5:5258.

31. Kossoy A, et al. (2015) Optical and structural properties of ultra-thin gold films. AdvOpt Mater 3:71–77.

32. Kwon KY, Sirowatka B, Weber A, Li W (2013) Opto- μECoG array: A hybrid neuralinterface with transparent μECoG electrode array and integrated LEDs for opto-genetics. IEEE Trans Biomed Circuits Syst 7:593–600.

33. Kuzum D, et al. (2014) Transparent and flexible low noise graphene electrodes forsimultaneous electrophysiology and neuroimaging. Nat Commun 5:5259.

34. Yao J, et al. (2011) Electrical and photosensitive characteristics of a-IGZO TFTs relatedto oxygen vacancy. IEEE Trans Electron Dev 58:1121–1126.

35. Milvich J, et al. (2015) Flexible low-voltage organic phototransistors based on air-stabledinaphtho [ 2, 3-b : 2 0, 3 0 -f ] thieno [ 3, 2-b ] thiophene (DNTT).Org Electron 20:63–68.

36. Vosgueritchian M, Lipomi DJ, Bao Z (2012) Highly conductive and transparent PEDOT:PSS films with a fluorosurfactant for stretchable and flexible transparent electrodes.Adv Funct Mater 22:421–428.

37. Wang DH, Kyaw AKK, Gupta V, Bazan GC, Heeger AJ (2013) Enhanced efficiencyparameters of solution-processable small-molecule solar cells depending on ito sheetresistance. Adv Energy Mater 3:1161–1165.

38. Kaltenbrunner M, et al. (2013) An ultra-lightweight design for imperceptible plasticelectronics. Nature 499:458–463.

39. Richner TJ, et al. (2014) Optogenetic micro-electrocorticography for modulating andlocalizing cerebral cortex activity. J Neural Eng 11:016010.

40. Fang H, et al. (2016) Ultrathin, transferred layers of thermally grown silicon dioxide asbiofluid barriers for biointegrated flexible electronic systems. Proc Natl Acad Sci USA113:11682–11687.

41. Ji ZG, et al. (2012) Light-evoked somatosensory perception of transgenic rats thatexpress channelrhodopsin-2 in dorsal root ganglion cells. PLoS One 7:e32699.

42. Ishizuka T, Kakuda M, Araki R, Yawo H (2006) Kinetic evaluation of photosensitivity in genet-ically engineered neurons expressing green algae light-gated channels. Neurosci Res 54:85–94.

43. Boyden ES, Zhang F, Bamberg E, Nagel G, Deisseroth K (2005) Millisecond-timescale,genetically targeted optical control of neural activity. Nat Neurosci 8:1263–1268.

44. Gu X, Yao C, Liu Y, Hsing IM (2016) 16-Channel organic electrochemical transistor array forin vitro conduction mapping of cardiac action potential. Adv Healthc Mater 5:2345–2351.

45. Someya T, et al. (2005) Conformable, flexible, large-area networks of pressure and thermalsensors with organic transistor active matrixes. Proc Natl Acad Sci USA 102:12321–12325.

46. Jeuris K, Groenendaal L, Verheyen H, Louwet F, De Schryver FC (2003) Light stabilityof 3,4-ethylenedioxythiophene-based derivatives. Synth Met 132:289–295.

47. Khurram A, Seymour JP (2013) Investigation of the photoelectrochemical effect inoptoelectrodes and potential uses for implantable electrode characterization. ConfProc IEEE Eng Med Biol Soc 2013:3032–3035.

48. Ludwig KA, Uram JD, Yang J, Martin DC, Kipke DR (2006) Chronic neural recordingsusing silicon microelectrode arrays electrochemically deposited with a poly(3,4-eth-ylenedioxythiophene) (PEDOT) film. J Neural Eng 3:59–70.

49. Braendlein M, et al. (2017) Lactate detection in tumor cell cultures using organictransistor circuits. Adv Mater 29:1605744.

Lee et al. PNAS | October 3, 2017 | vol. 114 | no. 40 | 10559

APP

LIED

PHYS

ICAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Dec

embe

r 8,

202

0