Available online at www.sciencedirect.com ScienceDirect Novel transducers for high-channel-count neuroelectronic recording interfaces Anton Guimerà-Brunet1,2, Eduard Masvidal-Codina1,2, Jose Cisneros-Fernández1, Francesc Serra-Graells1,3 and Jose A Garrido4,5 Neuroelectronic interfaces with the nervous system are an scales, its study is particularly challenging. At microscale, essential technology in state-of-the-art neuroscience research high spatial resolution is needed to resolve single unit aiming to uncover the fundamental working mechanisms of the activity to capture local information processing. At mac- brain. Progress towards increased spatio-temporal resolution roscale, it is required to cover a large brain area to detect has been tightly linked to the advance of microelectronics neural dynamics among different brain regions. At tem- technology and novel materials. Translation of these poral level, synaptic transmission and spiking activity technologies to neuroscience has resulted in multichannel takes place on the timescale from milliseconds to seconds, neural probes and acquisition systems enabling the recording while changes in the activity patterns related to learning, of brain signals using thousands of channels. This review memory or disease ranges from seconds to months or provides an overview of state-of-the-art neuroelectronic years. Technologies intended to monitor brain activity technologies, with emphasis on recording site architectures must therefore span over these wide ranges of temporal which enable the implementation of addressable arrays for and spatial scales, and also take into account the variabil- high-channel-count neural interfaces. In this field, active ity among species and individuals [1,2]. transduction mechanisms are gaining importance fueled by novel materials, as they facilitate the implementation of high Through the years, neuroscience has benefited from density addressable arrays. advances in a wide variety of research fields to develop neurotechnological tools able to tackle these outstanding Addresses 1 challenges, combining information from different tech- Institut de Microelectrònica de Barcelona, IMB-CNM (CSIC), Esfera niques to cover a wide range of scales [3]. Despite that UAB, Bellaterra, Spain large areas of the brain can be observed non-invasively 2 Centro de Investigación Biomédica en Red en Bioingenierı́a, with indirect measurements of the brain’s activity via Biomateriales y Nanomedicina (CIBER-BBN), Madrid, Spain 3 imaging techniques (fMRI or PET), the electrical basis of Universitat Autònoma de Barcelona, Spain the brain activity makes electrophysiology of paramount 4 Catalan Institute of Nanoscience and Nanotechnology (ICN2), CSIC and The Barcelona Institute of Science and Technology (BIST), Campus importance for both fundamental and clinical applications UAB, Bellaterra, Barcelona, Spain [4,5]. Electrical interfaces with the nervous system have 5 Institució Catalana de Recerca i Estudis Avançats (ICREA), Barcelona, been used consistently to monitor neural activity at Spain different spatial resolutions and invasiveness. Non-inva- sively, electroencephalography (EEG) is widely used in Current Opinion in Biotechnology 2021, 72:39–47 the clinical practice; however, the attenuation of the recorded signals by the skull limits its spatial resolution, This review comes from a themed issue on Tissue, cell and pathway engineering requiring from more invasive neural interfaces to obtain relevant information such as electrocorticography Edited by Mahsa Shoaran and Stéphanie P Lacour (ECoG) or intracortical implants inserted in the cortex For a complete overview see the Issue and the Editorial or deeper structures. Available online 22nd October 2021 https://doi.org/10.1016/j.copbio.2021.10.002 Currently, improvements in neural interface technology 0958-1669/ã 2021 The Author(s). Published by Elsevier Ltd. This is an can be grouped in three main directions [6–8]. One major open access article under the CC BY-NC-ND license (http://creative- line of research is focused on minimizing the foreign body commons.org/licenses/by-nc-nd/4.0/). response which limits the long-term functionality of neural interfaces [9,10]. A second line deals with increas- ing the number of recording sites, either for higher Introduction density or for larger brain area coverage. A major bottle- Revealing the fundamental working mechanisms of the neck for scaling up channel numbers is the connectivity of brain represents one of the biggest scientific challenges, electrodes to acquisition circuits. Finally, another major with deep intellectual significance as well as huge clinical challenge of neuroelectronic interfaces is the transduction value for the treatment of neurological disorders. Since efficiency to capture the neural signals with high fidelity brain activity spans over multiple spatial and temporal and signal-to-noise ratio over the broad frequency www.sciencedirect.com Current Opinion in Biotechnology 2021, 72:39–47 40 Tissue, cell and pathway engineering bandwidth of brain activity when downscaling the record- Single addressed passive transducers ing area. Historically, passive electrodes have been the most com- monly used transducers for recording brain activity. Its Taking a neural interface system-engineering perspec- progress has been closely related to improvements of tive, in this review, we focus on recent developments of microelectronic technology, which has enabled the evo- novel transduction mechanisms for high-fidelity, large- lution from simple microwire electrodes to modern mir- area and high spatial resolution recording of brain activ- croelectrode arrays (MEAs) [15]. The transducers’ per- ity. Active devices based on transistors are promising formance depends on the electrode-tissue interface, transducers to solve the challenge of connectivity that wherein the case of passive electrodes, its complex arises for high-channel count neural interfaces: they impedance is the main characteristic for recording appli- facilitate the implementation of addressable and multi- cations [16]. In the literature, the impedance modulus at plexed arrays, hence reducing the connectivity con- 1 kHz is often reported as a figure of merit to describe the strains and allowing interfaces with thousands of record- electrode-tissue interface. However, as the electrode’s ing sites. intrinsic noise is associated to the real component of the electrode-tissue impedance, the integration of this com- ponent over the recording bandwidth could be proposed Transducer technologies as a more informative parameter [17]. Generally, the The activity of neuronal cells involves ionic trans-mem- neural signal recorded with an electrode is amplified by brane currents that generate extracellular field potentials. a differential amplifier, forming a voltage divider between Thus, any electrophysiology method requires an efficient the electrode impedance (Ze) and the input impedance of transducer to convert ionic-generated field potentials into the amplifier (Za) (Figure 1b). This configuration can lead an electronic signal. To complete the recording system, a to a non-negligible frequency-dependent attenuation first signal conditioning is performed by an analog-front- when the Ze << Za condition is not fulfilled end (AFE), and then an analog-to-digital converter (Figure 1d) [18]. In addition, high impedance values also (ADC) translates the neural signals into digital data increase the cross-talk among channels [12]. Since the streams (Figure 1a). The performance of the full acquisi- electrode impedance is inversely proportional to its area, tion chain can be characterized by the signal-to-noise ratio the downscale capabilities provided by micro-fabrication (SNR), the signal-to-distortion ratio (SDR), and the cross- techniques are limited by the increasing electrode imped- talk between channels [11,12,13]. As transducers work ance with decreasing electrode size. To overcome this at the interface between the ionic charge carries of limitation, most research efforts have been focused in the biological tissue and the solid-state charge carries of development of electrode materials with higher capaci- the acquisition system, their electrical properties are of tance per unit area [19]. One approach is to engineer paramount importance for a high-fidelity recording. Nota- materials to become porous, increasing their surface area bly, the AFE also plays a significant role in the acquisition and thus their capacitance for the same recording area chain since it is the first element in contact with the [20,21]. Recently, the use of biocompatible platinum transducer; the AFE provides a first amplification step but nanorods (PtNR) to implement thousands of recording can also introduce noise and non-idealities that could sites has allowed mapping the complex temporal dynam- severely affect the quality of the recorded signal. Since ics from the cortex [22]. Alternatively, organic materials the operation of the transducer and the AFE are tightly such as PEDOT:PSS are widely used as electrode coating coupled, it is convenient to discuss these two components to obtain low electrode impedance values [23,24]. In this together. case, the ions from brain tissue can diffuse within the electrode material, which result in a volume-dependent To rationalize the capabilities of micro-fabrication tech- capacitance [25,26]. niques for downscaling the recording sites and increasing their density [1,2], the optimal electrode size and array Passive transducers are susceptible to DC offsets and low- configuration should be designed according to applica- frequency drifts present at the electrode-tissue interface tion-specific requirements. In particular, for detecting due to, for instance, slow electrochemical reactions and and distinguishing extracellular action potentials (EAPs) ion accumulation, impairing high fidelity recording of of nearby neurons, smaller electrodes (diameter < brain’s infra-slow activity and introducing high common 20 mm) should be used to minimize the spatial-averaging voltage offsets [27,28]. This limitation precluded the use effect [14]. For local field potentials and infra-slow of DC-coupled AFEs, and has promoted the use of high- signals, on the other hand, large electrodes can offer a pass filter (HPF) architectures [29] (Figure 1b). The low superior performance in terms of noise. However, the frequency corner of HPF (0.1 Hz) leaves the infra-slow electrode size should be balanced with the required signals outside of the acquisition frequency band. Despite spatial resolution for determining the source of the achieving low-frequency DC blocking and well defined signals, for example, to perform current source density signal bandwidth acquisition, at such frequencies the analysis [11]. flicker noise from the CMOS amplifier can dominate over Current Opinion in Biotechnology 2021, 72:39–47 www.sciencedirect.com High-channel-count recording interfaces Guimerà-Brunet et al. 41 Figure 1 (a) Signal Transducer AFE Multiplexer Analog-to-digital Computer source converter b Mux ADC b (b) Passive (c) Active Vref Rg Vsig Ids Vin Vsig Electrode I/V Vref Vout Vout Rs VGS Vg Rs Vsig Cg-e Vs Vd Re-e Ce-e Vin Ids Idc Vsig ˙gm Vsig Ids Vout Rds Ids [A] Z’ e Zӏe Zӏa Source Drain dIds Channel gm= Vin dVgs VDs Vgs [V] (d) (e) Vgs Vsig High Z′e 108 Low Z′e ӏ OSFET OECT gSGFET ⎥Z⎥ [Ω] Za 106 mS gm [S] 104 μS Dielectric 10 mHz 10 Hz 10 kHz 1.0 Frequency [Hz] Channel Gain (Vin/Vsig ) 0.8 Non-flat Attenuation Non-linearity - > Harmonic distortion -> Gain = Z′ a/ (Z′a + Z′e) freq. response Phase shifts 0.6 Ids [A] Distortion[%] gm / gm DC 0.4 1 0 10¯ 2 10¯ 1 100 101 102 103 104 10 mHz 10 Hz 10 kHz Frequency [Hz] Vgs [V] Frequency [Hz] (f) Signal coupling Noise Downscaling Biasing Passive Z′ a/ (Z′a + Z′e) Z′e α 1/A N.A. gm A independent vdz, vgs Active gm gm Irms α 1/A Current Opinion in Biotechnology Recording electrophysiological signals with active and passive transducers. (a): Recording chain schematic including the signal source, the transducer, the analog front-end (AFE), the anti-aliasing filter, the multiplexer (Mux), an analog-to-digital covnerter (ADC) and a computer. (b)–(f): Comparison of passive and active transduction. (b)–(c) left: Cross-section of the interface between the tissue and the transducer and superimposed electrical schematic. (b)–(c) right: electrical schematic of the transducer and first amplification stage. For passive transducers recordings is based on differential amplifiers (schematic shown in top) and signal is coupled through a voltage divider composed by the electrode (Z’e) and amplifier (Z’a) impedances. For active transducers recording requires device biasing (Vds,Vgs) and recording the modulated current signals. Signal coupling can be approximated by the derivative of the transfer function at the operation point (gm) (d): Impedance spectra for high and low electrode impedance and frequency response of the recording gain for each case. (e): Top. Cros-section of the mechanisms of different transistor types and comparison of the transconductance magnitude and frequency response. OSFET: Oxide-semiconductor field-effect transistor; OECT: Organic electrochemical transistor. gSGFET: graphene solution-gated field-effect transistor. Bottom: Non idealities such as harmonic distortion due to non-linear transfer curve and non-flat transconductance frequency response should be taken into account to achieve high-fidelity recordings. (f): Table comparing the scaling and dependence of the main figures of merit of passive and active transducers. the brain signal. Therefore, noise mitigation techniques Neurotechnologies aimed to record neural activity have including chopping and correlated-double sampling strat- also profited from the increased integration density of egies are required to ensure electronic noise values lower microelectronics. This evolution has allowed to improve than intrinsic electrode noise [30]. the AFEs and the integration of several channels together www.sciencedirect.com Current Opinion in Biotechnology 2021, 72:39–47 42 Tissue, cell and pathway engineering with digitizing circuitry into a single chip [31]. This by switching with a given periodicity to define the HPF integration, together with the availability of wireless cut-off frequency (Figure 2b). technology, has simplified the experimental preparations, enabling data recordings in freely moving animal experi- Other materials such as organic polymers or carbon-based ments under more realistic behavioral conditions [32–34]. materials can be used in direct contact with the brain tissue, resulting in larger transconductance values. Along Active transducers facilitate high density addressable this line, PEDOT:PSS has been used to implement the arrays so-called Organic Electrochemical Transistors (OECTs). The adoption of multiplexing or addressable strategies for Despite the low carrier mobility of PEDOT:PSS, large increasing the channel count requires the incorporation of transconductance values can be obtained thanks to its switching elements (Figure 2a, b) [2]. These elements volumetric capacitance [44,45]. Other research has introduce parasitic components in the signal path, explored carbon-based materials for active neural trans- decreasing the input impedance and therefore ducers. Taking advantage of graphene’s high carrier compromising the Ze << Za condition for high density mobility, chemical stability and field-effect properties arrays. Alternative to passive transducers, active transdu- [39], graphene-based solution-gated field-effect transis- cers based on transistors provide a local signal amplifica- tors (gSGFETs) have been used in direct contact with the tion thus minimizing the impact of parasitic elements. brain tissue [46,47]. gSGFETs are able to overcome the Moreover, they can be arranged in row-column address- limitations of passive electrodes in DC-coupled opera- able arrays, simplifying and reducing the connectivity tion. Graphene’s chemical stability significantly reduces [35,36,37]. In active transducers, the extracellular field signal drift, and the signal coupling through the interfacial potential modulates the conductivity of the channel capacitance warrants a stable transconductance even at material, which is electrically contacted by two metal very low frequencies [48]. This capability has led to the contacts (drain and source) resulting in a three terminal recent demonstration of high-resolution mapping of infra- device with a transistor configuration. The third terminal slow brain signals with gSGFETs [34,49]. (gate) is electrically contacted through the brain tissue with a reference electrode (Figure 1c). The efficiency of Regarding signal quality, it is important to note that in the modulation of the drain-source current by the gate active transducers the transconductance might be non- voltage is given by the so-called transconductance (gm), linear and frequency-dependent (Figure 1e), limiting the which depends on the mobility of charge carriers in the fidelity of the transduction if not taken into account channel and the interfacial (channel-tissue) capacitance. [13,50]. Importantly, the transistor’s transconductance The chemical stability of the channel material in contact is not dependent on the channel area but on its aspect with the brain tissue plays an important role in the ratio (width/length), which allows reducing the transistor implementation of active transducers, resulting in differ- dimensions without impairing the transconductance per- ent device configurations [38,39] (Figure 1e). formance. However, the major factor limiting transistor downscaling is its intrinsic noise, which is inversely Some semiconductor material, such as silicon, require the proportional to the active area. The most representative use of an insulation layer to prevent unwanted electro- figure of merit of noise is the gate voltage equivalent chemical reactions. However, benefiting from CMOS noise, defined as the integral over the operation band- technology maturity, passive electrodes are connected width of the current noise divided by the transconduc- to the gate of silicon transistors for implementing the tance (Figure 1f) [51,52]. so-called extended gate configuration. In this way, high integration density applications have been developed taking profit from the local amplification provided by Challenges of read-out electronics for high the silicon transistor to build active sensing pixels channel count recording interfaces [32,37,40,41]. However, the materials used to imple- Two technology trends can be clearly defined in the effort ment the passive electrodes also suffer from the previ- towards the development of high-channel-count transdu- ously described DC offsets and drifts that can compro- cers, typically based on monolithic or hybrid technology mise the system performance. Decoupling of these solutions. In the case of monolithic solutions, the full unwanted drifts requires the use of a HPF, which increase acquisition chain, from the transducer to the digitizer, is the area of the active pixel. For that, different approaches built in the same device. Based on CMOS technology, have been used to minimize the area required for imple- this approach has resulted in important advances for menting this HPF. For instance, the use of an input intracortical recordings on the pursuit of high spatial capacitance (Figure 2a, b), together with resistances act resolution, enabling the identification of the activity of as a HPF and help to set the transducer operating point thousands of single neurons [40,41]. On the other hand, [32,41]. Another design approach that requires less area in hybrid solutions, the transducers are built separately is based on the use of auto-zero structures based on a from the acquisition electronics [35,37,53]. This switching transistor [40,42,43], which can reset any drift approach has been favored in applications aiming to cover Current Opinion in Biotechnology 2021, 72:39–47 www.sciencedirect.com High-channel-count recording interfaces Guimerà-Brunet et al. 43 Figure 2 Recording Sites Peripheral Electronics Active Element Summary References (a) SADD<0> SADD<1> SADD<2> Pixel Group SADD 2T 1R 1C Type: Monolithic Addressable [32] Vout Coupling: AC [41] Connectivity: m x n + Supplies Vout<0> [54] № Channels: m x n Mux ADC VREF Recording channels: m [55] SADD<3> SADD<4> SADD b (b) SCOL<0> SCOL<1> SCOL Column Control SCOL 3T Vout<0> Type: Monolithic Time-Multiplex Coupling: DC-Auto Zero [40] SCAL Connectivity: m x n + Supplies [42] VREF № Channels: m x n Vout<1> Vrow m× Mux ADC b b 2T 1C Type: Hybrid Time-Multiplex Vout Coupling: AC [37] Connectivity: m + n + Supplies № Channels: m x n ×n Vrow (c) SCOL<0> SCOL<1> SCOL VCOL Irow<0> Vrow<0> b 1T ADC Type: Hybrid Time-Multiplex [36] Coupling: DC [53] Irow<1> Vrow<1> Connectivity: m + n m× b № Channels: m x n [61] ADC Irow Irow Vrow b ADC ×n (d) VCOL<0> VCOL<1> VCOL Freq VCOL Irow<0> Vrow<0> b 1T ADC Type: Hybrid Frequency-Multiplex Coupling: DC [35] Vrow<1> Connectivity: m + n m× Irow<1> ADC b № Channels: m x n Irow Irow Vrow b ADC ×n Non-Recording Site Recording Site T Transistor C Capacitor R Resistor Current Opinion in Biotechnology Acquisition architectures for addressable arrays based on active transducers. (a): Addressable array with (m  n) available channels without multiplexing capabilities, where only (m) configurable channels can be recorded simultaneously. Where the anti-aliasing filter is located before the ADC minimizing the out of band noise folding. (b) and (c): Time domain multiplexing array (m  n) with reduced connectivity (m + n), where active sites are scanned sequentially (The sampling frequency for each row is Fs  n, where Fs is the channel sampling frequency). (b) Top: Monolithic 3T active element with auto-zero integration to achieve AC coupling with better area efficiency than RC structures. (b) Bottom: Hybrid silicon integration on flexible substrates, with AC coupling (1C), on-site pre-amplification and column switch (2T). (c): Current mode DC-coupled active element, based on solution gated transistors (1T) with column switches placed outside the flexible array. (d): Frequency domain multiplexing array (m  n) without switches, where all sites are active at same time, and frequency band division scheme for each column. large areas of the brain, enabling the mapping of brain reach high integration densities, having a dedicated functional circuits and rhythms across brain regions. amplifier for each recording site is not efficient. This has motivated the adoption of active pixel structures The growth of monolithic solutions has been powered by inherited from CMOS image sensors [43], allocating the increasing capabilities of CMOS fabrication techni- the resources not needed for the signal transduction far ques which allows integration of signal transducers in the away from the recording area. To enable this reallocation, same fabrication process. To optimize the area usage and the connectivity with the recording area should be www.sciencedirect.com Current Opinion in Biotechnology 2021, 72:39–47 44 Tissue, cell and pathway engineering simplified. For that, the arrangement of recording sites in integrated into flexible polyimide substrates has allowed addressable arrays has been widely used for reducing the to implement up to thousand recording sites with only 64 number of connections: for a matrix of m x n recording (36 + 28) connections by using a TDM read-out strategy sites (where m and n are the number of columns and rows, [37,58]; however, at the cost of a very complex integra- respectively), the number of connections is m + n and the tion. Alternatively, the use of active transducers compati- number of required AFEs is m (Figure 2). As commented ble with flexible substrates can simplify the technology before, the use of active transducers has played an impor- integration. In this line, OECT based on PEDOT:PSS tant role for implementing the addressable arrays, mini- have been explored to implement TDM by mixing an mizing the pixel area while preserving the signal integrity active transducer with an on-site switch in small arrays [2]. (Figure 2c) [59]. This approach was improved by the use of switchless arrays and enhanced conformability [36,60]. While the adoption of addressable arrays improves the Taking advantage of the high carrier mobility in gra- resources allocation, the signal read-out of these arrays phene, arrays of gSGFETs have recently been used to presents some bottlenecks, for which the requirements of implement switchless arrays with TDM [61]. The impact the target application should guide the adopted solution. of the tracks resistance on the signal quality and cross-talk Mainly, the signal of all sensors in the addressable array should be considered. This is particularly relevant in the can be acquired simultaneously by using multiplexing case of OECT and gSGFETs transducers, as the output approaches; alternatively, the addressing capabilities impedance is lower than the extended gate implementa- could be used to configure a selection of sensors to be tions, thus being more sensitive to parasitic track resis- read-out. The difference between both approaches relies tances [53]. Moreover, the capability of gSGFETs to act in the location of the anti-aliasing filter into the acquisi- as a signal mixer enables the study of amplitude modula- tion chain, and its impact on both the sampling rate and tion techniques for implementing frequency division noise performance [12]. Importantly, in the case of time multiplexing (FDM) strategies, avoiding the aliasing division multiplexing (TDM), when all sensors are problems presented by TDM strategies (Figure 2d) recorded simultaneously, the sampling rate of the multi- [35]. In both TDM and FDM gSGFETs implementa- plexing system should be higher than the sampling rate tions, the wide-bandwidth recording capability of these for each individual sensor (sampling rate scales linearly devices was demonstrated, preserving high-fidelity with the number of columns). To avoid the out-of-band recording of infra-slow activity. noise folding inherent to TDM, the anti-aliasing filter should be placed before the switching elements. This means that it should be placed into the sensing pixel, Conclusions and outlook where area is more scarce. Even though, some efforts In the last decades, translation of microelectronics tech- have been done by implementing a current integration nology to neuroscience has resulted in integrated circuits with a certain frequency to build a low pass filtering at that enable monitoring brain activity with thousands of pixel level [54]. Summarizing, while simultaneous multi- recording sites. The increased spatial resolution provided plexing readings can be achieved at expenses of lower by monolithic solutions, together with the simultaneous sampling rate and higher noise [40,54,55], addressable use of multiple shanks have revealed very detailed local configurations are preferred to achieve a performance information from different areas of the brain [40,41]. On similar to that of point to point systems [41,56]; in the other hand, hybrid solutions used in mECoG record- the case of the addressable configurations, the increased ings aim to cover large areas of the brain, allowing to map integration density provides certain degree of freedom to brain functional circuits and rhythms [22,34,58], select the sensors to record. where active transducers compatible with flexible sub- strates are highly suited building blocks for high-channel- In contrast to the monolithic designs described above, count flexible interfaces. However, despite the significant hybrid solutions are based on building separately the progress, there is no established a technology able to acquisition electronics from the array of transducers, cover the wide range of spatial and temporal scales of taking advantage of recent advances in novel materials the brain activity, requiring a choose according to the for flexible electronics technology [8,38]. In contrast to requirements of the target application. Translation of the use of the rather invasive monolithic shanks, mECoG high-channel-count recording technology to human recordings on top of the cortex are able to provide applications could provide many clinical benefits as also information from large brain areas in a less invasive allow improved decoding efficiency in brain-machine way [57]. The good performance of mECoG arrays relies interfaces. However, with the number of recording chan- on the substrate conformability to the brain’s surface nels progressively increasing in the coming years, we morphology, for which rigid monolithic solutions are foresee a shift from the technology-governed bottleneck not suitable. Hybrid interfaces can also employ address- of the number of recording sites to a bottleneck governed able array configurations to reduce the connectivity foot- by data management, its transmission, storage, analysis print. For instance, the use of doped silicon nanoribbons and interpretation [62,63]. Current Opinion in Biotechnology 2021, 72:39–47 www.sciencedirect.com High-channel-count recording interfaces Guimerà-Brunet et al. 45 Conflict of interest statement 14. Harris KD, Quiroga RQ, Freeman J, Smith SL: Improving data quality in neuronal population recordings. Nat Neurosci 2016, Nothing declared. 19:1165-1174. 15. Obaid A, Hanna M-E, Wu Y-W, Kollo M, Racz R, Angle MR, Müller J, Brackbill N, Wray W, Franke F et al.: Massively parallel Acknowledgements microwire arrays integrated with CMOS chips for neural recording. Sci Adv 2020, 6:eaay2789. This work has been funded by the European Union’s Horizon 2020 research and innovation program under Grant Agreement No. 732032(BrainCom) 16. Boehler C, Carli S, Fadiga L, Stieglitz T, Asplund M: Tutorial: and Grant Agreement No. 881603 (Graphene Flagship) and co-funded by guidelines for standardized performance tests for electrodes the European Regional Development Funds (ERDF) allocated to the intended for neural interfaces and bioelectronics. Nat Protoc Programa operatiu FEDER de Catalunya 2014–2020, with the support of 2020, 15:3557-3578 http://dx.doi.org/10.1038/s41596-020-0389- the Secretaria d’Universitats i Recerca of the Departament d’Empresa i 2. Coneixement of the Generalitat de Catalunya for emerging technology clusters devoted to the valorization and transfer of research results 17. Mierzejewski M, Steins H, Kshirsagar P, Jones PD: The noise  and impedance of microelectrodes. J Neural Eng 2020, (GraphCAT 001-P-001702). The ICN2 is supported by the Severo Ochoa 17:052001. Centres of Excellence program, funded by the Spanish Research Agency (AEI, grant no. SEV-2017-0706), and by the CERCA Program/Generalitat 18. Nelson MJ, Pouget P, Nilsen EA, Patten CD, Schall JD: Review of de Catalunya. signal distortion through metal microelectrode recording circuits and filters. J Neurosci Methods 2008, 169:141-157. 19. Choi JS, Lee HJ, Rajaraman S, Kim D-H: Recent advances in References and recommended reading  three-dimensional microelectrode array technologies for in Papers of particular interest, published within the period of review, vitro and in vivo cardiac and neuronal interfaces. Biosens have been highlighted as: Bioelectron 2021, 171:112687.  of special interest 20. Boehler C, Vieira DM, Egert U, Asplund M: NanoPt—a  of outstanding interest nanostructured electrode coating for neural recording and microstimulation. ACS Appl Mater Interfaces 2020, 12:14855- 14865. 1. Nurmikko A: Challenges for large-scale cortical interfaces.  Neuron 2020, 108:259-269. 21. Ganji M, Paulk AC, Yang JC, Vahidi NW, Lee SH, Liu R, Hossain L, 2. Obien MEJ, Deligkaris K, Bullmann T, Bakkum DJ, Frey U: Arneodo EM, Thunemann M, Shigyo M et al.: Selective Revealing neuronal function through microelectrode array formation of porous pt nanorods for highly electrochemically recordings. Front Neurosci 2015, 8. efficient neural electrode interfaces. Nano Lett 2019, 19:6244- 6254. 3. Frank JA, Antonini M-J, Anikeeva P: Next-generation interfaces for studying neural function. Nat Biotechnol 2019, 37:1013- 22. Tchoe Y, Bourhis AM, Cleary DR, Stedelin B, Lee J, Tonsfeldt KJ, 1023.  Brown EC, Siler D, Paulk AC, Yang JC et al.: Human brain mapping with multi-thousand channel PtNRGrids resolves 4. Emmenegger V, Obien MEJ, Franke F, Hierlemann A: novel spatiotemporal dynamics. arXiv 2021:210309206 Technologies to study action potential propagation with a Mapping of cortical brain activity with thousands of electrodes, showing focus on HD-MEAs. Front Cell Neurosci 2019, 13. the capability of high-channel-count recording interfaces to resolve complex temporal dynamics from the cortical surface. 5. Jastrzebska-Perfect P, Chowdhury S, Spyropoulos GD, Zhao Z, Cea C, Gelinas JN, Khodagholy D: Translational 23. Jones PD, Moskalyuk A, Barthold C, Gutöhrlein K, Heusel G, neuroelectronics. Adv Funct Mater 2020, 30:1909165. Schröppel B, Samba R, Giugliano M: Low-impedance 3D PEDOT:PSS ultramicroelectrodes. Front Neurosci 2020, 14. 6. Fattahi P, Yang G, Kim G, Abidian MR: A review of organic and inorganic biomaterials for neural interfaces. Adv Mater 2014, 24. Khodagholy D, Gelinas JN, Thesen T, Doyle W, Devinsky O, 26:1846-1885. Malliaras GG, Buzsáki G: NeuroGrid: recording action potentials from the surface of the brain. Nat Neurosci 2015, 7. Hong G, Yang X, Zhou T, Lieber CM: Mesh electronics: a new 18:310-315. paradigm for tissue-like brain probes. Curr Opin Neurobiol 2018, 50:33-41. 25. Volkov AV, Wijeratne K, Mitraka E, Ail U, Zhao D, Tybrandt K, Andreasen JW, Berggren M, Crispin X, Zozoulenko IV: 8. Woods GA, Rommelfanger NJ, Hong G: Bioinspired materials for Understanding the capacitance of PEDOT:PSS. Adv Funct in vivo bioelectronic neural interfaces. Matter 2020, 3:1087- Mater 2017, 27:1700329. 1113. 9. Wellman SM, Eles JR, Ludwig KA, Seymour JP, Michelson NJ, 26. Donahue MJ, Sanchez-Sanchez A, Inal S, Qu J, Owens RM, McFadden WE, Vazquez AL, Kozai TDY: A materials roadmap to Mecerreyes D, Malliaras GG, Martin DC: Tailoring PEDOT functional neural interface design. Adv Funct Mater 2018, properties for applications in bioelectronics. Mater Sci Eng R 28:1701269. Rep 2020, 140:100546. 10. Yang W, Gong Y, Li W: A review: electrode and packaging 27. Chandrakumar H, Markovic D: A high dynamic-range neural  materials for neurophysiology recording implants. Front recording chopper amplifier for simultaneous neural Bioeng Biotechnol 2021, 8. recording and stimulation. IEEE J Solid-State Circuits 2017, 52:645-656. 11. Viswam V, Obien MEJ, Franke F, Frey U, Hierlemann A: Optimal  electrode size for multi-scale extracellular-potential recording 28. Li C, Narayan RK, Wu P-M, Rajan N, Wu Z, Mehan N, Golanov EV, from neuronal assemblies. Front Neurosci 2019, 13:385 Ahn CH, Hartings JA: Evaluation of microelectrode materials for This paper studies both by simulations and with experimental data, the direct-current electrocorticography. J Neural Eng 2015, effects that influence signal detection and guide transducer size design 13:016008. for a given application. 29. Harrison RR: A low-power, low-noise CMOS amplifier for 12. Pérez-Prieto N, Delgado-Restituto M: Recording strategies for neural recording applications. In 2002 IEEE International high channel count, densely spaced microelectrode arrays. Symposium on Circuits and Systems. Proceedings (Cat. Front Neurosci 2021, 15:869. No.02CH53). 2002:V. 13. Garcia-Cortadella R, Masvidal-Codina E, la Cruz JMD, Schäfer N, 30. Wu R, Huijsing JH, Makinwa KAA: Dynamic offset cancellation Schwesig G, Jeschke C, Martinez-Aguilar J, Sanchez-Vives MV, techniques for operational amplifiers. In Precision Villa R, Illa X et al.: Distortion-free sensing of neural activity Instrumentation Amplifiers and Read-Out Integrated Circuits. using graphene transistors. Small 2020, 16:1906640. Edited by Wu R, Huijsing JH, Makinwa KAA. Springer; 2013:21-49. www.sciencedirect.com Current Opinion in Biotechnology 2021, 72:39–47 46 Tissue, cell and pathway engineering 31. Harrison RR: The design of integrated circuits to observe brain Sanchez-Vives MV: Mapping brain activity with flexible activity. Proc IEEE 2008, 96:1203-1216. graphene micro-transistors. 2D Materials 2017, 4:025040. 32. Jun JJ, Steinmetz NA, Siegle JH, Denman DJ, Bauza M, 47. Hébert C, Masvidal-Codina E, Suarez-Perez A, Calia AB, Barbarits B, Lee AK, Anastassiou CA, Andrei A, Aydın Ç et al.: Fully Piret G, Garcia-Cortadella R, Illa X, Garcia EDC, De la Cruz integrated silicon probes for high-density recording of neural Sanchez JM, Casals DV et al.: Flexible graphene solution- activity. Nature 2017, 551:232-236. gated field-effect transistors: efficient transducers for micro-electrocorticography. Adv Funct Mater 2018, 33. Pazhouhandeh MR, O’Leary G, Weisspapir I, Groppe D, Nguyen X, 28:1703976. Abdelhalim K, Jafari HM, Valiante TA, Carlen P, Verma N et al.: 22.8 adaptively clock-boosted auto-ranging responsive  T, 48. Masvidal-Codina E, Illa X, Dasilva M, Calia AB, Dragojevi c neurostimulator for emerging neuromodulation applications.  Vidal-Rosas EE, Prats-Alfonso E, Martı́nez-Aguilar J, De la 2019 IEEE International Solid-State Circuits Conference - (ISSCC) Cruz Sanchez JM, Garcia-Cortadella R et al.: High- 2019:374-376. resolution mapping of infraslow cortical brain activity enabled by graphene microtransistors. Nat Mater 2019, 34. Garcia-Cortadella R, Schwesig G, Jeschke C, Illa X, Gray AL, 18:280-288. Savage S, Stamatidou E, Schiessl I, Masvidal-Codina E, Kostarelos K et al.: Graphene active sensor arrays for long-term 49. Masvidal-Codina E, Smith TM, Rathore D, Gao Y, Illa X, Prats- and wireless mapping of wide frequency band epicortical Alfonso E, Corro ED, Calia AB, Rius G, Martin-Fernandez I et al.: brain activity. Nat Commun 2021, 12:211. Characterization of optogenetically-induced cortical spreading depression in awake mice using graphene micro- 35. Garcia-Cortadella R, Schäfer N, Cisneros-Fernandez J, Ré L, Illa X, transistor arrays. J Neural Eng 2021, 18:055002.  Schwesig G, Moya A, Santiago S, Guirado G, Villa R et al.: Switchless multiplexing of graphene active sensor arrays for 50. Mackin C, McVay E, Palacios T: Frequency response of brain mapping. Nano Lett 2020, 20:3528-3537 graphene electrolyte-gated field-effect transistors. Sensors This paper demonstrates the capability of graphene transistors to imple- 2018, 18:494. ment a frequency domain multiplexing without switches. 51. Polyravas AG, Curto VF, Schaefer N, Calia AB, Guimera-Brunet A, 36. Lee W, Kim D, Matsuhisa N, Nagase M, Sekino M, Malliaras GG, Garrido JA, Malliaras GG: Impact of contact overlap on Yokota T, Someya T: Transparent, conformable, active transconductance, noise in organic electrochemical multielectrode array using organic electrochemical transistors. Flex Print Electron 2019, 4:044003. transistors. Proc Natl Acad Sci U S A 2017, 114:10554-10559. 52. Mavredakis N, Garcia Cortadella R, Illa X, Schaefer N, Bonaccini 37. Viventi J, Kim D-H, Vigeland L, Frechette ES, Blanco JA, Kim Y-S, Calia A, Anton-Guimerà-Brunet A, Garrido J, Jiménez D: Bias Avrin AE, Tiruvadi VR, Hwang S-W, Vanleer AC et al.: Flexible, dependent variability of low-frequency noise in single-layer foldable, actively multiplexed, high-density electrode array graphene FETs. Nanosc Adv 2020, 2:5450-5460. for mapping brain activity in vivo. Nat Neurosci 2011, 14:1599- 1605. 53. Schaefer N, Garcia-Cortadella R, Martı́nez-Aguilar J, Schwesig G, Illa X, Lara AM, Santiago S, Hébert C, Guirado G, Villa R et al.: 38. Song E, Li J, Won SM, Bai W, Rogers JA: Materials for flexible Multiplexed neural sensor array of graphene solution-gated  bioelectronic systems as chronic neural interfaces. Nat Mater field-effect transistors. 2D Mater 2020, 7:025046. 2020, 19:590-603 Review on recording materials and encapsulations for flexible neural 54. Raducanu BC, Yazicioglu RF, Lopez CM, Ballini M, Putzeys J, interfaces. Wang S, Andrei A, Welkenhuysen M, van Helleputte N, Musa S et al.: Time multiplexed active neural probe with 678 parallel 39. Hess LH, Hauf MV, Seifert M, Speck F, Seyller T, Stutzmann M, recording sites. 2016 46th European Solid-State Device Research Sharp ID, Garrido JA: High-transconductance graphene Conference (ESSDERC) 2016:385-388. solution-gated field effect transistors. Appl Phys Lett 2011, 99:033503. 55. Lopez CM, Putzeys J, Raducanu BC, Ballini M, Wang S, Andrei A, Rochus V, Vandebriel R, Severi S, Hoof CV et al.: A neural probe 40. Boi F, Perentos N, Lecomte A, Schwesig G, Zordan S, Sirota A, with up to 966 electrodes and up to 384 configurable channels Berdondini L, Angotzi GN: Multi-shanks SiNAPS active pixel in 0.13 mm SOI CMOS. IEEE Trans Biomed Circuits Syst 2017, sensor CMOS probe: 1024 simultaneously recording channels 11:510-522. for high-density intracortical brain mapping. bioRxiv 2020:749911 http://dx.doi.org/10.1101/749911. 56. Dragas J, Viswam V, Shadmani A, Chen Y, Bounik R, Stettler A, Radivojevic M, Geissler S, Obien MEJ, Müller J et al.: In vitro 41. Steinmetz NA, Aydin C, Lebedeva A, Okun M, Pachitariu M, multi-functional microelectrode array featuring 59  Bauza M, Beau M, Bhagat J, Böhm C, Broux M et al.: Neuropixels 760 electrodes, 2048 electrophysiology channels, 2.0: a miniaturized high-density probe for stable, long-term stimulation, impedance measurement, and neurotransmitter brain recordings. Science 2021, 372:eabf4588 detection channels. IEEE J Solid-State Circuits 2017, 52:1576- Monolithic implementation of thousands of recording sites distributed in 1590. four shanks. 57. Anumanchipalli GK, Chartier J, Chang EF: Speech synthesis from 42. Angotzi GN, Boi F, Lecomte A, Miele E, Malerba M, Zucca S, neural decoding of spoken sentences. Nature 2019, 568:493-  Casile A, Berdondini L: SiNAPS: an implantable active pixel 498. sensor CMOS-probe for simultaneous large-scale neural recordings. Biosens Bioelectron 2019, 126:355-364. 58. Chiang C-H, Won SM, Orsborn AL, Yu KJ, Trumpis M, Bent B,  Wang C, Xue Y, Min S, Woods V et al.: Development of a neural 43. Berdondini L, Imfeld K, Maccione A, Tedesco M, Neukom S, interface for high-definition, long-term recording in rodents Koudelka-Hep M, Martinoia S: Active pixel sensor array for high and nonhuman primates. Sci Transl Med 2020, 12 spatio-temporal resolution electrophysiological recordings Recording of brain activity from the cortex with a time domain read-out of from single cell to large scale neuronal networks. Lab Chip thousand recording sites and long-term stability evaluation. 2009, 9:2644-2651. 59. Lee W, Kim D, Rivnay J, Matsuhisa N, Lonjaret T, Yokota T, 44. Keene ST, van der Pol TPA, Zakhidov D, Weijtens CHL, Yawo H, Sekino M, Malliaras GG, Someya T: Integration of Janssen RAJ, Salleo A, van de Burgt Y: Enhancement-mode organic electrochemical and field-effect transistors for PEDOT:PSS organic electrochemical transistors using ultraflexible, high temporal resolution electrophysiology molecular de-doping. Adv Mater 2020, 32:2000270. arrays. Adv Mater 2016, 28:9722-9728. 45. Rivnay J, Inal S, Salleo A, Owens RM, Berggren M, Malliaras GG: 60. Lee W, Kobayashi S, Nagase M, Jimbo Y, Saito I, Inoue Y, Organic electrochemical transistors. Nat Rev Mater 2018, 3:1- Yambe T, Sekino M, Malliaras GG, Yokota T et al.: 14. Nonthrombogenic, stretchable, active multielectrode 46. Blaschke BM, Tort-Colet N, Guimerà-Brunet A, Weinert J, array for electroanatomical mapping. Sci Adv 2018, 4: Rousseau L, Heimann A, Drieschner S, Kempski O, Villa R, eaau2426. Current Opinion in Biotechnology 2021, 72:39–47 www.sciencedirect.com High-channel-count recording interfaces Guimerà-Brunet et al. 47 61. Cisneros-Fernandez J, Garcia-Cortadella R, Illa X, Martinez- 62. Stevenson IH, Kording KP: How advances in neural recording Aguilar J, Paetzold J, Mohrlok R, Kurnoth M, Jeschke C, Teres L, affect data analysis. Nat Neurosci 2011, 14:139-142. Garrido JA et al.: A 1024-channel 10-bit 36-W/ch CMOS ROIC for multiplexed GFET-only sensor arrays in brain mapping. 63. Paninski L, Cunningham J: Neural data science: accelerating IEEE Trans Biomed Circuits Syst 2021:3113556 http://dx.doi.org/ the experiment-analysis-theory cycle in large-scale 10.1109/TBCAS.2021.3113556. neuroscience. Curr Opin Neurobiol 2018, 50:232-241. www.sciencedirect.com Current Opinion in Biotechnology 2021, 72:39–47