Recent developments and future perspectives on neuroelectronic devices

: Neuroscientific discoveries and the development of recording and stimulation tools are deeply connected. Over the past decades, the progress in seamlessly integrating such tools in the form of neuroelectronic devices has been tremendous. Here, we review recent advances and key aspects of this goal. Firstly, we illustrate improvements with respect to the coupling between cells/tissue and recording/ stimulation electrodes. Thereafter, we cover attempts to mitigate the foreign body response by reducing the devices ’ invasiveness. We follow up with a description of specialized electronic hardware aimed at the needs of bioelectronic applications. Lastly, we outline how additional modalities such as optical techniques or ultrasound could in the future be integrated into neuroelectronic implants. Anwendungen. Zuletzt zeigen wir auf, wie neue Modalitäten z.B. durch optische Techniken oder Ultraschall zukünftig in neuroelektronische Implantate integriert werden könnten. Elektrische Ableitung; Stimulation;


Introduction
Since its very beginning, neuroscience has been deeply intertwined with methods and technologies of measuring or controlling the electrical activity in biological tissue. Ultimately, this intricate connection can be traced back to the founding experiments of Luigi Galvani on frog legs (Galvani, 1791). In the 1950s, the development of tungsten microelectrodes enabled recordings from single neurons in living animals (Hubel, 1957). Continuing the push toward higher resolution, Neher and Sakmann introduced the patch-clamp technique in the 1970s, which allowed characterizing individual ion channels (Neher and Sakmann, 1976). These developments have reshaped neuroscience since the middle of the 20th century triggering an ever-accelerating technological development. In particular over the past decades, new microfabrication approaches, as well as dedicated efforts on merging biology and artificial probes, have led to a cornucopia of new tools. Probes to interrogate the nervous systems are constantly improving in terms of signal quality, number of channels, and seamlessness of their integration (Won et al., 2018). At the same time, the accompanying electronics are becoming smaller and require less energy (van Dongen and Serdijn, 2016) while additional modalities such as optical or chemical stimulation are added.
This review aims to provide an overview of recent developments as well as the core challenges met when integrating artificial devices with the nervous system or biological systems in general. While the main focus is on technologies to be used in in vivo applications, a few examples for in vitro applications are discussed as well to demonstrate specific approaches in this domain. In doing so, we do not aim for a complete historical overview.
Instead, we hope to provide the reader with an insight into the central aspects and most recent advances in the field of bio-and neuroelectronics. Overall, the review is structured in four sections that mirror the key aspects toward which efforts have been directed over the past years (compare Figure 1). First, we will illustrate improvements in the coupling between the electrode and the tissue. Thereafter, approaches to reduce the devices' invasiveness for better long-term applicability will be described. Next, we will outline technological progress with regards to detection principles and the underlying electronics of the devices. The last section will shortly highlight novel approaches to include other-than-electrical principles (e.g. optical or fluidic technologies) into neuroelectronic devices before we conclude the review.

Electrode-tissue coupling
The vast majority of techniques are concerned with recording or eliciting electrical signals, which is rooted in the critical role that electrical activity plays in the nervous system. In this context, the interface between an electrode and the cells or tissue of interest takes a central place. The properties of this interface are dominated by several factors. In particular, the interface impedance contributes to the attenuation of signals that are supposed to be recorded Figure 1: Schematic illustrating the four sectors in which neuroelectronic devices have advanced over the past years. Improvements in tissueelectrode coupling (top left; scale bars correspond to 2, 250, 500, and 200 µm, top-left to bottom-right, respectively) yield better signal transduction. A reduction in the devices' invasiveness (top right; scale bars correspond to 25 µm, approx. 10 mm, and 200 µm, top-left to bottom-right, respectively) mitigates the foreign body response. Specialized hardware (bottom left) delivers higher sensitivity and reduced power and spatial requirements for the accompanying electronics. The introduction of new modalities for recording and stimulation (bottom right) expands the neuroelectronic toolset. Details and copyright information for all four subpanels can be found in Figures 2,3,4,and 5. by the electrode or likewise to be elicited in the tissue. At the same time, the impedance is a crucial factor in the voltage noise observed when recording with the electrode. On a general level, a lower impedance will mean better signal transduction and lower noise. Another factor is the so-called seal or junction resistance. This resistance represents undesired electrical pathways between the electrode and the ground. Increasing the seal resistance by tightening the tissue-electrode connection will generally result in improved signal fidelity.

Modifying the interface impedance
In the simplest case, the interface impedance can be approximated as an RC element. As such, the majority of interface modifications target either an increase in capacitance (e.g. by increasing the effective interface area) or a decrease in the interface resistance (e.g. by facilitating Faradaic charge transfer). A typical example is the functionalization of electrodes with platinum black. Since improvements in the adhesion of platinum black were made in the late 1980s (Marrese, 1987), it has been a predominant method to decrease the interface impedance by roughening the surface. Even despite known cytotoxic side effects (Schuettler et al., 2005), improvements in platinum black coating continue to pose a viable alternative for impedance reduction (Boehler et al., 2015;compare Figure  2a and b). Similar to the deposition of platinum black, electroplated gold was shown to be able to decrease the interface impedance by increased surface roughness (Brüggemann et al., 2011;Cui and Martin, 2003a). Aside from modifications to the surface geometry, chemical modifications have also been shown to bear great benefit in optimizing the interface for signal transduction and noise reduction Ferro and Melosh, 2018). Among these, the application of conductive polymers takes a special place. The charge transport in such materials is often a mixture of electron/hole and ionic mechanisms. Applied to the electrodes of a neuroelectronic device, they can greatly improve the interface properties (Inal et al., 2018;Liang et al., 2021). Consequently, conductive polymers have found numerous applications in bioelectronic settings (Bettucci et al., 2021) with a continuously increasing toolset for structuring them on the microscale (Zhang and Travas-Sejdic, 2021).
While a number of conductive polymers have been applied in modifying bioelectronic interfaces, poly(3,4-ethylenedioxythiophene) (PEDOT), in particular when doped with polystyrene sulfonate (PSS), is considered by many to be the gold standard in this context (Liang et al., 2021). Already early on, the fuzzy nature of electrodeposited PEDOT:PSS films was not only shown to be beneficial for the electrode impedance but also for cell growth in vitro (Cui and Martin, 2003b). In addition to its benefits in a recording context, PEDOT was also shown to be an excellent choice for modifying stimulation electrodes. Similar to recording scenarios, the low interfacial impedance is beneficial for signal transduction across the interface. In this context, it was found that PEDOT-modified electrodes exhibit excellent charge storage capacities (Wilks et al., 2009). This property describes the amount of charge that can be applied through an electrode before undesired electrochemical reactions take place. As such, an increased charge storage capacity helps mitigate the increased current density requirements resulting from electrode miniaturization. Despite its long-standing history in this field, PEDOT:PSS has not lost its relevance as demonstrated by its recent application in conjunction with conductive hydrogels (Ferlauto et al., 2018) or in the realization of stretchable peripheral nerve interfaces (Decataldo et al., 2019).
Naturally, both approaches of impedance reduction can be combined, i.e. the application of conductive polymers can be tuned in order to increase their surface area to ultimately lower the overall impedance. Specifically, conductive polymers can be deposited in ways that render them porous (Yang and Martin, 2004a, b) or in the form of nanotubes that offer an increased surface area due to their low dimensionality (Abidian and Martin, 2008;. Such nanotube-decorated electrodes have been shown to also promote improved adhesion of cells to the electrode (Abidian et al., 2010), an effect that was also observed when coating electrodes with blends of conductive polymers and biomolecules (Cui et al., 2001;Kim et al., 2007).

Tightening the tissue-electrode connection
As illustrated by these last examples, the degree of contact between the tissue and the electrode plays a similarly crucial role when probing biological tissue. Conceptually, a tighter coupling to the target tissue results in a higher seal or junction resistance and thereby more efficient transduction of signals both when recording or stimulating. Consequently, an interface as described above with cells closely interacting with the electrode's pores or nanostructures is expected to improve the electrode performance. However, the modification of the interface does not have to be performed before the addition of the cells. Instead, a tight interface can also be created by electropolymerizing PEDOT directly in the intercellular space both in cell culture studies and in living tissue explants (Richardsony-Burns et al., 2007a, b). Effectively, such approaches try to integrate the electrode directly into the target tissue's natural structure. Vice versa, a variety of methods use structural motives that the cells spontaneously engulf, i.e. they result in the tissue integrating more tightly with the electrode. Since the nanotechnology boom of the 2000s a large toolset to generate structures similar in size to those naturally encountered by cells is available. Numerous examples of how such structures can be used to create a tight tissue-electrode connection have been presented ranging from nanoparticles over nanopillars to nanocones and nanomushrooms (Angle et al., 2015;Spira and Hai, 2013). The majority of these techniques aim at tightening the contact between the electrode and the cell body. As such, they are to a certain degree focussed on applications in the central nervous system (CNS). While a number of these approaches will also bear advantage in applications involving the peripheral nervous system (PNS), there are also dedicated techniques that capitalize on properties specific to the PNS. One such property is given by the cylindrical nature of most structures found in the PNS. Exploiting this symmetry to create a more intricate tissue-electrode seal, microfabricated clip systems that embed electrodes for recording and stimulation have recently been presented (Lissandrello et al., 2017;Otchy et al., 2020; see Figure 2c and d). Other approaches make use of, for instance, adhesive hydrogel patches that wrap around the electrode and the nerve (Forssell et al., 2019;Horn et al., 2021;Ong et al., 2018).

Reduced invasiveness
The previous section highlighted the crucial importance of a tight coupling between the neuroelectronic device and the tissue of interest. In most cases, however, this coupling is required to last over days, if not months, in order to conduct biologically or clinically meaningful studies. Such timescales open up a new challenge as they allow activation of the body's defense mechanismsthe 'foreign body response' (FBR). Two of the main drivers of the FBR are suspected to be adsorption-related conformation changes in proteins (Hu et al., 2001) and a mechanical mismatch between the implanted materials and the target tissue (Moshayedi et al., 2014). The result is a cascade of events that attacks and encapsulates the implant leading to device failure (Barrese et al., 2013;Xie et al., 2014) or loss of neurons in the implant's vicinity (Biran et al., 2005). To counteract the FBR, neuroprobes are required to smoothly integrate into the soft, wet, and constantly moving environment of the body . The following section will thus illustrate strategies to mitigate FBR-related decreases in probe performance. In doing so, we will describe approaches that use modifications of the probe surface, soft materials to reduce mechanical mismatch, and overall reduced size to cause as little trauma in the target tissue as possible.

Surface modification
As described above, the probe surface can have a strong influence on the degree of FBR observed upon device implantation. Thus, modification of the device surface is an attractive candidate for effective mitigation strategies. In this context, not only a variety of both chemical but also structural approaches allow tuning of the cell-surface interaction (Stevens and George, 2005). Due to their soft and biologically inspired nature, hydrogels lend themselves to this application (Kim et al., 2010). Recent developments even yielded photostructurable gelatin systems that could be integrated into standard fabrication processes (Kang et al., 2020). While beneficial for cell adhesion, such coatings increase the electrode-cell distance and thereby lower the seal resistance. To mitigate this aspect, conducting polymers can be electropolymerized inside of the hydrogel. The thereby obtained structure of the conducting polymers even decreases the electrode impedance potentially making up for the loss in seal resistance (Kim et al., 2004). Besides hydrogels, a number of bio-derived materials such as silk, collagen, chitin, or cellulose have already been demonstrated as viable substrates for bioelectronic interfaces (Pradhan et al., 2020; see Figure 3b). Increasingly, progress in the processability of these materials is blurring the line between classical microfabrication and bio-derived materials (Ju et al., 2020;compare Figure 3a). An approach that renders this line even less clear is based on so-called biohybrid interfaces. Early examples of this approach demonstrated that stem cell-seeded implants show lower neuron loss and glial encapsulation immediately after implantation (Purcell et al., 2009). Since then, the use of cells to relay or amplify artificial electronic signals has been realized in numerous studies in vitro and in vivo targeting both the PNS and the CNS (Rochford et al., 2019).

Addressing the mechanical mismatch
As highlighted by Moshayedi et al. (2014), the mechanical properties of neuroelectronic devices are another important factor in how well they fare in long-term applications. In this context, two strategies are commonly applied. The first employs materials that exhibit low Young's moduli, i.e. show mechanical properties closer to those of biological tissue than classic silicon or metal materials. Here, bio-friendly or bio-derived substrate materials have been demonstrated (compare Figure 3c; Adly et al., 2018;Maiolo et al., 2019). A common problem, however, is posed by the fact that metals usually suffer from a strong dependence of their conductivity on the applied strain, i.e. bending or stretching of the device results in a strong increase in the resistance. To mitigate this aspect, mesh-like conductor networks made from e.g. metal nanowires or carbon nanotubes can be applied (Lienemann et al., 2021;Terkan et al., 2020;Tybrandt et al., 2018). Alternatively, the microstructure of gold films can be engineered so that they remain conductive even when exposed to notable strain rates (Lacour et al., 2003(Lacour et al., , 2006. Both approaches have been successfully implemented in the recording and stimulation of nerve activity (Lienemann et al., 2021;Minev et al., 2015). Another approach is represented by using elastic conductive materials (Giagka and Serdijn, 2018;Jeong et al., 2015;Rogers et al., 2010;Sunwoo et al., 2020), such as composites of PEDOT:PSS and polyurethane (Cuttaz et al., 2019) or liquid metal conductors (Dong et al., 2021). Recent advances in this context cover rapid prototyping of soft bioelectronic implants (Afanasenkau et al., 2020) and ultrasoft mesh structures that seem to pose no hindrance to the mechanical activity of cardiac cells (Lee et al., 2018; see Figure 3d).
The second strategy is based on the fact that absolute mechanical stiffness is not only dependent on a material's bulk properties but also on its size and geometry. Similar to how both fingernails and hair consist of keratin, thinning down a macroscopically "stiff" material can result in highly pliable structures. As device thicknesses in the range of only a few micrometers are not uncommon in this area, materials that are sufficiently robust to allow handling and insertion are needed. Two of the most common examples in this category are polyimide (Borda et al., 2020;Kireev et al., 2019;Sperry et al., 2018) and parylene (Khodagholy et al., 2011(Khodagholy et al., , 2013(Khodagholy et al., , 2015, both of which offer additional benefits in terms of chemical inertness and biocompatibility. To increase the number of channels while maintaining a small form factor, multilayered and multiplexed devices have been proposed (Leccardi et al., 2019;Viventi et al., 2011). Overall, technological progress in this context has achieved a stage where even signals of individual cells can be detected (Almasri et al., 2020;Khodagholy et al., 2015), and transparent devices allowing parallel optogenetic modulation (Lee et al., 2017) have become possible. In terms of overall technology readiness, NeuraLink has recently presented a platform that performs robotic implantation of and wireless recordings from thousands of channels in living animals (Musk and Neuralink, 2019). In particular, recordings from pigs and proof-of-concept brain-computer interface scenarios using monkeys have been demonstrated. Lastly, apart from thin probes, the importance of probe geometry in reducing glial scaring was shown early on (Seymour and Kipke, 2007). Since then, this approach has been heavily explored to yield shuttle techniques that only leave thin implants within the tissue (Williamson et al., 2015), injectable micron-scale meshes (Zhou et al., 2017), carbon fiber arrays (Jiman et al., 2020), and highly parallelized microwire bundles coupled to complementary metal-oxide semiconductor (CMOS) chips Obaid et al., 2020).

Hardware developments
So far, we have covered the progress neuroelectronic devices have made in increasing the efficiency of signal transduction as well as preventing the body from rejecting them. However, before such signals can be interpreted by a researcher they have to be both amplified and digitized. Here, several aspects of importance will be illustrated in the following section. Firstly, principles of local amplification to prevent long leads from introducing noise will be described. Subsequently, we will briefly dive into specialized electronics that aim to keep both spatial and energy requirements at a minimum. Lastly, wireless approaches to minimize constraints during e.g. behavioral studies will be introduced.

Local amplification
Despite the aforementioned advances in improving transduction, biological signals remain low in amplitude in the majority of scenarios. At the same time, they commonly have to be routed over significant lengths before they can be captured by recording electronics. For the most part, this combination results in challenging signal-to-noise ratios at the point of digitization. To address this challenge, different approaches aim to amplify the signal before routingand thereby before the introduction of noise. A prominent approach to achieve such a local amplification is to use fieldeffect transistors (FETs) as recording elements. Bioelectronic applications of FETs have been presented not only on the basis of both PEDOT:PSS  but also twodimensional materials such as graphene (compare Figure  4a; Garcia-Cortadella et al., 2020;Schaefer et al., 2020). Alternatively, it was shown that it is even possible to directly integrate the necessary amplification circuitry within the recording device (Frey et al., 2009). The fabrication complexity needed to accomplish this is currently only manageable with CMOS fabrication. At the same time, however, CMOS allows ultra-dense electrode arrays that enable trakking action potentials down to a subcellular resolution (Bakkum et al., 2013;Lewandowska et al., 2015). More recently, CMOS fabricated shank electrodes that bring a similar electrode density to in vivo applications have been presented (Boi et al., 2020;Sayed Herbawi et al., 2018; see Figure 4b).

Specialized electronics
Despite the sizeable challenges of detecting biological signals, the electronic backend needed to digitize and record the data often represents a major bottleneck. This can have a number of reasons starting from the sheer amount of data to be handled to size or energy constraints when considering wireless, implantable systems. Driven by these constraints  , 2018). c) A specialized analog to digital converter consuming as little as 0.5 µW (© 2013 IEEE. Adapted, with permission, from Li et al., 2013). d) Circuit diagram for safe (i.e. charge-balanced) and power-efficient stimulation hardware (© 2020 IEEE. Adapted, with permission, from Guan et al., 2020). and the increasing demand in terms of the functionality of neuroelectronic devices, the development of specialized hardware for bioelectronic recordings is a vivid field of research. In particular, the combined demands of recording, stimulation, and ideally pre-processing of the signals are of central interest Rincón Montes et al., 2019). Especially when considering non-tethered, mobile applications, power consumption is a crucial factor. Here, digitization elements requiring only µW of power ; see Figure 4d), low-energy signal processing elements (Haddad and Serdijn, 2009;Hiseni et al., 2009), or efficient stimulation circuitry (compare Figure 4c;van Dongen and Serdijn, 2016;Guan et al., 2020;Kolovou Kouri et al., 2021) have been proposed. In particular for the latter, development has reached a point where even syringe-injectable stimulation hardware is within reach (Li et al., 2015).

New recording/stimulation modalities
Up to this point, the main focus of most applications we discussed revolved around purely electrical recording and stimulation. While electrical signals areparticularly with respect to the nervous systemof unquestionable importance, a range of other physical phenomena can be recorded from biological systems or used to elicit biological signals (Rivnay et al., 2017). Examples of such multimodal recordings include, for instance, the detection of strain and/or temperature in biological tissue (Xu et al., 2014;Yokota et al., 2015). Apart from the measurement of such physical quantities within the tissue, chemical signals belong to the body's own communication system. Here, protocols that counteract the common problem of electrode fouling have recently been presented to allow longterm monitoring of biologically relevant electroactive species in vivo (Weltin et al., 2019). In order to deliver chemical stimuli, passive release from the aforementioned conductive polymer nanotubes was investigated ). In applications that require a more discrete delivery of chemicals, electrophoretic (Proctor et al., 2019;Simon et al., 2009), electrochemical (Boehler et al., 2017), or microfluidic (Guo et al., 2021) delivery were shown to be viable options. In order to directly measure the tissue response to the stimulus, even chemical stimulation and electrical sensing with the very same electrode were demonstrated (Jonsson et al., 2016;compare Figure 5b).
The spread of optogenetics over the past two decades has demonstrated how versatile of a tool optical light can be. Naturally, implantable optoelectronic devices that allow coupling between the techniques described above and the potential of optogenetics are thus of great interest (Pisanello, 2019;Pisanello et al., 2016). But apart from delivering stimuli, optical tools can also be used to record photometric data in freely behaving animals (Pisano et al., 2019;refer to Figure 5a). Similar to electrical devices, optical technology was also advanced to reach wireless Samineni et al., 2017) or injectable (Kim et al., 2013; see Figure 5c) levels. Even without genetic modification, the use of photocapacitive or photofaradaic (i.e. materials that convert light into electricity) allows harnessing the tether-free nature of optical stimulation (Ghezzi et al., 2011;Paltrinieri et al., 2021;Rand et al., 2018;compare Figure 5d). Similarly, even ultrasonic or magnetoelectric signal transduction has been discussed (Seo et al., 2016;Singer et al., 2020).

Conclusions
Over the past 20 years, the key barriers for efficient multimodal recording and seamless integration of neuroelectronics devices have been identified and put under scrutinous investigation. Although some of these barriers have proven tougher than expected, immense progress was made to the point where clinical applications of more Figure 5: New modalities to broaden the neuroelectronic toolset. a) Tapered fibers can be used in depth-resolved photometric recordings (adpated by permission from Pisano et al., 2019). b) Schematic of a bioelectronic neural pixel, a device that can record electrically and deliver chemical stimuli in the same position (Jonsson et al., 2016). c) Image sequence of the insertion of an injectable, cellular scale optoelectronic device (From Kim et al., 2013. Reprinted with permission from AAAS). d) Schematic of direct photocapacitive stimulation. Shining light onto the stimulation site results in polarization of the photocapacitor, which in turn depolarizes adjacent cells (Rand et al., 2018). sophisticated bioelectric approaches are within reach. Minev et al., for instance, were able to use soft neuroelectronic implants to restore locomotion after paralyzing injury in rats (Minev et al., 2015). They were able to demonstrate that their thin, soft implant not only limits inflammation but also results in a more natural gait after recovery. Similarly, the improved device-tissue coupling offered by hydrogel-supported electrode arrays allows fiber-selective stimulation of the vagus nerve in rats while presumably limiting compression (Forssell et al., 2019;Horn et al., 2021). In the CNS, organic electrochemical transistor arrays can record dopamine levels in different positions over time. Such recordings recently revealed a complex cross-talk between mesolimbic and nigrostriatal pathways in the rat brain (Xie et al., 2020).
We believe that in particular the concerted efforts from different fields such as electronic engineering, biochemistry, material science, and microfabrication ensure steady progress toward further blurring the line between technology and biology. Once this line is passed, neuroelectronic devices will deliver on their promise for many applications in prosthetic, diagnostic, and therapeutic applications.
Author contributions: All the authors have accepted responsibility for the entire content of this submitted manuscript and approved submission. Research funding: None declared. Conflict of interest statement: The authors declare no conflicts of interest regarding this article.