Article | Published:

# Rational design of silicon structures for optically controlled multiscale biointerfaces

## Abstract

Silicon-based materials have been widely used in biological applications. However, remotely controlled and interconnect-free silicon configurations have been rarely explored, because of limited fundamental understanding of the complex physicochemical processes that occur at interfaces between silicon and biological materials. Here, we describe rational design principles, guided by biology, for establishing intracellular, intercellular and extracellular silicon-based interfaces, where the silicon and the biological targets have matched properties. We focused on light-induced processes at these interfaces, and developed a set of matrices to quantify and differentiate the capacitive, Faradaic and thermal outputs from about 30 different silicon materials in saline. We show that these interfaces are useful for the light-controlled non-genetic modulation of intracellular calcium dynamics, of cytoskeletal structures and transport, of cellular excitability, of neurotransmitter release from brain slices and of brain activity in vivo.

## Main

Fundamental discoveries of new forms and new properties in materials can lead to new designs of biophysical tools and biomedical devices1,2,3,4,5,6,7,8,9,10,11,12,13,14,15. For example, dopant-modulated and kinked silicon (Si) nanowires allow for intracellular electrical recording from cardiomyocytes with a field-effect-transistor configuration2. Bendable integrated circuits, based on Si nanoscale membranes and their seamless interface with a thermal oxide, open the way for long-lived bioelectronic implants for the heart6. Although the electrically registered device components have yielded impressive results, remotely controlled and freestanding systems are rarely employed in biointerface studies16,17,18,19,20,21,22,23. This is largely due to our limited understanding of the physicochemical processes at the freestanding material surfaces under physiological conditions. In particular, a quantitative understanding of the light-induced electrical, electrochemical and thermal pathways across multiple length scales, if achieved, would likely promote future biointerface innovations.

Here, we formulate a rational design principle for a series of Si-based freestanding ‘biotronics’ with length scales from nanometre to centimetre, which establish intra-, inter- and extracellular biointerfaces. The organization of the paper follows this order (Supplementary Fig. 1). First, we introduce a biology-guided Si-based biomaterial design, which first considers the material structures and mechanics and then the efficient signal transductions at the Si surfaces in saline. Next, we recommend three classes of materials for establishing biointerfaces across different length scales. Finally, we demonstrate the utility of these new devices by showing light-controlled non-genetic modulations of intracellular calcium dynamics, cytoskeleton-based transport and structures, cellular excitability, neurotransmitter release from brain slices, and brain activities in a mouse model.

## The principle of biology-guided biointerface design

Si displays many size- and doping-dependent physicochemical processes. To efficiently leverage these processes in the context of biointerfaces, the Si-based materials or devices should be in tight contact (Fig. 1a, Selection I) with their biological counterparts. Such tight interfaces can be established by protein-associated tethering and active motions at the organelle level, by dynamic cellular focal adhesions at the single cell and tissue level, and by van der Waals forces at the organ level. To promote these forces, we focus our Si materials on nanowire geometries (at the organelle level)24, membranes with rough surfaces (at the cell and tissue level)25, and flexible and distributed meshes (at the organ level)1, where at least one dimension of the material properties can be tuned to promote tight interfaces (Fig. 1b). After the material/device structures are determined, we are next in a position to examine the effects of other orthogonal controls (for example, size, doping, surface chemistry) to produce the desirable physicochemical processes (Fig. 1a, Selection II) at the biointerfaces. These two-step selections, guided by the need to form tight junctions (Selection I) and efficient signal transductions (Selection II) with the biological targets, would narrow the material options to those that are better suited for the targeted biophysical or biomedical questions.

## Selection for material structures

For example, to enable intracellular biointerfaces, we chose Si nanowires as recent studies showed that they can be internalized into mammalian cells through phagocytosis24 and, when inside, form active interfaces with cytoskeletal systems14. In addition, to promote light absorption from single nanowire structures26, we deposited nanocrystalline Si shells over a thin, vapour–liquid–solid (VLS)-grown Si nanowire backbone (~50 nm in diameter; Fig. 1c, left and upper right). Cross-sectional and side-view transmission electron microscopy (TEM) images reveal that >95% of the total volume is nanocrystalline (Fig. 1c, lower right).

To build extracellular interfaces with single cells or small tissues, we can examine planar Si structures with uniformly doped or dopant-modulated configurations to identify the effect of doping. In particular, a p-type–intrinsic–n-type (p–i–n) Si diode junction was synthesized by chemical vapour deposition (CVD) of intrinsic and n-type Si layers (~140 and ~190 nm in thickness, respectively) over a p-type Si semiconductor-on-insulator (SOI) substrate (p-type Si thickness, ~2 µm; Supplementary Fig. 2). Cross-sectional (scanning) TEM images taken at the interface between the SOI wafer and the as-deposited layers indicate a columnar shell structure with a sharp and oxide-free interface (<1 nm junction width; Fig. 1d, left and upper right). While the p-type substrate is single crystalline, the i-/n-layers are nanocrystalline (Fig. 1d, lower right, and Supplementary Fig. 2), which is reminiscent of the Si nanostructures used for thin-film solar cells. The surface of the nanocrystalline layer is rough, which would promote cellular focal adhesions25. Beyond doping controls, we also prepared metal (that is, gold, silver and platinum) nanoparticle-covered Si diodes by electroless deposition to expand the repertoire of Si-based biointerfaces (Supplementary Figs. 35).

Finally, to create a conformal interface with a soft and curvilinear organ, for example, a mouse brain cortex, we explored a flexible device made of a distributed mesh of Si membrane (~2.3 μm in thickness) and a porous polydimethylsiloxane (PDMS) substrate (~120 μm in thickness; Fig. 1e and Supplementary Fig. 6). The holey structures in both Si and PDMS can mitigate the stress accumulated across a large device area (Supplementary Fig. 7) and enhance the device mechanical compliance.

## Selection for material functions

Given that cellular physiology can be altered with an approximately picoampere-level ionic current, we next utilized a high-precision electrochemical tool, that is, a patch-clamp setup, to investigate the light-induced and biointerface-relevant physicochemical processes (Figs. 1a, Selection II, and 2; Methods) that originated from the freestanding Si surfaces27. Briefly, we immersed different types of Si materials (for example, nanocrystalline nanowires, and dopant-modulated, surface-treated and size-tuned nanomembranes, as selected from the first step) into a phosphate-buffered saline (PBS) solution and positioned glass micropipette electrodes in close proximity to the Si surfaces (~2 µm) where ionic flows across the pipette tips were measured in the voltage-clamp mode (Fig. 2a, upper, and Supplementary Fig. 8a). Specifically, we delivered light pulses (530 nm light-emitting diode (LED) or 532 nm laser, 10 ms) through a microscope objective to illuminate Si and recorded the ionic current dynamics under different pipette holding potentials (Fig. 2a, lower; Methods).

Using pulsed light illuminations, we developed a universal analysis (Methods, ‘Analysis of the photoresponse measurements’) and unambiguously identified and mostly importantly, decoupled two explicit and one implicit element of the photoresponses (Fig. 2b,c and Methods). In a representative trial from an Au-decorated p–i–n Si membrane (Fig. 2b, upper), we first noticed two ‘spiky’ features under a LED light pulse (~12.05 mW, ~500 μm spot size) with a power density of ~6 W cm−2. The upward (~86 nA, with a transient peak current density at the pipette tip of ~2,700 mA cm−2) and downward (~−34 nA, transient current density of ~1,100 mA cm−2) components correspond to capacitive charging/discharging processes at the Si/electrolyte interface (Fig. 2d, upper left, 1). The second photoresponse element is manifested as a long-lasting current with a lower amplitude (for example, ~2 nA for the same Au-decorated Si membrane; Fig. 2b, upper inset), which is indicative of a Faradaic current leading to redox reactions (Fig. 2d, upper left, 2). Metal-free p–i–n Si membranes only display symmetrical capacitive current spikes (Supplementary Fig. 9), with negligible Faradaic components detected from the local patch-clamp electrode. The last photoresponse element is implicit and it corresponds to the local temperature elevation of the solution due to the photothermal effect from Si. In this scenario, the nonradiative recombination of carriers converts part of the input photon energy into the vibrational energy of the Si lattice (Fig. 2d, lower left, 3), which dissipates heat through both Si and the surrounding electrolyte. Because the glass micropipette resistance is temperature dependent, we determined the thermal dynamics by fitting the recorded patch-clamp currents at various holding potentials (Methods). For example, we recorded an ~5.4 K peak temperature increase from a nanocrystalline Si nanowire on laser illumination (~47.1 mW, ~5 μm spot size) at ~240 kW cm−2 for 10 ms (Fig. 2b, lower).

To build quantitative matrices for these three photoresponse elements, we screened a library of Si-based materials to evaluate the impact of doping, surface chemistry and size (Fig. 3). We extracted the capacitive, Faradaic and thermal components from the patch-clamp recordings of 16 representative Si samples (Fig. 3) and projected them onto three axes to decouple any individual contributions (Fig. 2c). We first considered the simplest single-crystalline p-type Si SOI substrate (device layer thickness: ~2 µm; Fig. 2c, left). A typical experiment (LED illumination, ~12.05 mW, ~500 μm spot size, ~6 W cm−2) shows a small capacitive current of ~50 pA, and a transient peak current density of ~1.6 mA cm−2 at the pipette tip. However, on deposition of the intrinsic and n-type layers (that is, forming a p–i–n diode junction; Supplementary Figs. 9 and 10), the recorded capacitive current and transient peak current density were boosted to ~7,400 pA and ~235 mA cm−2, respectively (Fig. 2c, left, and Supplementary Figs. 9 and 10). This significant enhancement in the capacitive component is likely due to the enhanced light absorption from the nanocrystalline layers and more efficient charge separation by the built-in electric fields across the p–i–n diode junction (Fig. 2d, upper left). The polarity of the onset capacitive currents (that is, upward at the onset of light illumination) in both cases stays cathodic, although the dopant types of the electrolyte-interfacing layers are different (p in p-type SOI device layer, and n in p–i–n multilayers). This suggests that the primary light-generated carriers accumulated on the Si surfaces are electrons, which are the minority carriers in p-type SOI and the majority carriers in the p–i–n samples, respectively. This observation is similar to the device configurations used in traditional photoelectrochemical devices, that is, p-type semiconductors in contact with electrolytes would experience the band bending in such a manner that drives photogenerated electrons towards the p-type semiconductor/electrolyte interface, while the built-in electric fields in p–i–n devices sweep photogenerated electrons to the n-type semiconductor/electrolyte interface28. With a reversed doping sequence, both the n-type SOI substrate (~−48 pA) and the corresponding n–i–p (~−510 pA) diode junction display the opposite capacitive current polarity (Fig. 2c, left), as expected. Taken together, and consistent with the scenarios in traditional photoelectrochemical cells, the photocarriers that accumulate at the Si surfaces on light illumination are minority carriers in uniformly doped cases (that is, n- or p-type Si) and majority carriers when p–i–n or n–i–p junctions are formed. Finally, in these metal-free samples, the Faradaic and thermal components are negligible, for example, ~7 pA and ~0 K as peak values for the p–i–n multilayered sample (Supplementary Fig. 9, under ~6 W cm−2 LED illumination) so the dominant photoresponse element in metal-free Si membranes is the capacitive current.

As shown in the Au-decorated p–i–n Si membrane, the Faradaic current can reach ~2 nA (Fig. 2b, upper inset), suggesting a means of charge injection into the solution (Fig. 2d, upper left). We next explored multiple metals that were commonly exploited catalysts for photoelectrochemistry (for example, Au, Ag and Pt) by electroless deposition of nanoparticles onto p–i–n Si surfaces29 (Supplementary Figs. 35 and 1113). In all experiments, the introduction of metal species promoted both the capacitive and the Faradaic elements (Fig. 2c, middle, and Supplementary Figs. 11 and 13), with good stabilities over 1,000 repetitive illuminations (Supplementary Fig. 12), likely due to the fact that certain metals can more efficiently collect and solution-inject the photogenerated carriers (as opposed to carrier recombination in bulk Si). Among all the conditions tested, Au prepared by immersion of the p–i–n multilayered membrane in a 1 mM HAuCl4 solution yielded the highest capacitive (~86 nA) and Faradaic (~2 nA) currents (Fig. 2c, middle). The thermal components under ~6 W cm−2 LED illumination were negligible in all metal-decorated Si membranes.

Since single-cell or subcellular studies require highly localized interrogation, we measured the laser-induced (~47.1 mW, ~5 μm spot size, ~240 kW cm−2) photoresponses of Si materials with variable lateral dimensions. We observed decreased electrical/enhanced thermal elements with reduced sizes of p–i–n Si membranes (Fig. 2c, right, and Supplementary Fig. 14). In the case of intrinsic nanocrystalline Si nanowire, the nanoconfinement effect led to the highest photothermal response (~5.4 K peak temperature change) with negligible capacitive and Faradaic components (Fig. 2b, lower).

## Materials and devices for multiscale biointerfaces

Our physicochemical measurements (Fig. 1a, Selection II and Fig. 2) highlight p–i–n diode junction-enhanced capacitive currents, metal-enhanced capacitive and Faradaic currents, and nanoconfinement-enabled thermal responses (Figs. 2d, middle, and 3), all in freestanding configurations. For the present biointerface studies (Figs. 1a, lower right, 4 and 5), most Si materials such as a simple p-type Si membrane or p–i–n Si multilayers with small lateral dimensions (<500 μm), will not be considered given that their photoresponse components are small (Fig. 1a, excluded materials from Selection II, and Fig. 3, with lower colour intensity). We focused on the intrinsic nanocrystalline Si nanowires for intracellular and the related intercellular probing, and only used the photothermal effect. For single-cell or small-tissue-level inter- and extracellular studies, we primarily explored a light-induced capacitive effect, that is, we used a p–i–n Si multilayered membrane where the biological invasiveness from capacitive electrochemical currents are usually minimal30. Finally, given the organ-level modulation would require the strongest stimuli, we used the Au-coated p–i–n Si multilayered membrane for in vivo studies (Fig. 2d, right). Similar to the electrical stimulation of excitable tissues, where both capacitive and Faradaic currents take place at the electrode/electrolyte interface30, the Au-coated Si surfaces can deliver similar signals to the biological system for efficient tissue modulations.

## Organelle-level biointerfaces

We first considered Si nanowires for intracellular stimulation biointerfaces (Fig. 4) because it is an unexplored domain that is beyond the previously studied intracellular sensing or delivery. In a primary culture of neonatal rat dorsal root ganglia (DRG) and associated satellite glia, we noticed a cell-type-specific overlapping of nanocrystalline Si nanowires after ~24 h of co-culturing (by glial fibrillary acidic protein (GFAP)/NeuN staining in Fig. 4a and Supplementary Fig. 15; by S-100/Neurofilament staining in Supplementary Fig. 16). Statistical analysis of the nanowire-cell colocalization revealed that ~87% of total nanowires overlapped with glial cells, ~3% with neurons and ~10% stayed in the extracellular space (Fig. 4a, lower). Perinucleus clustering, rather than random intracellular distributions, of the colocalized nanowires suggests the internalization of these nanowires14,24. In addition, the presence of bent nanowires following the contours of a few glial cell membranes implies strong mechanical interactions between cells and nanowires14. As suggested by a recent study that label-free nanowires can be internalized through a phagocytosis pathway24, the fact that glial cells (versus neurons) do have phagocytic activities supports the observed selective glial internalization31. As a result, we studied the control of glial activities with internalized nanocrystalline nanowires as the remotely controlled stimulators. To this end, we illuminated an intracellularly bounded nanowire with a laser pulse (592 nm, ~14.4 mW, ~237 nm spot size, 1 ms) in the middle of a time-lapse calcium imaging series (Fig. 4b and Supplementary Figs. 8b and 17). On light illumination, the glial cell of interest, with the nanowire inside, experiences a fast calcium concentration increase followed by a slow decay. As the cell is being stimulated intracellularly, the observed calcium dynamics are likely related to the release of calcium from internal storage organelles, for example, endoplasmic reticulum and mitochondria, rather than the calcium influx through ion channels at the plasma membrane, and therefore may be extended to other non-excitable cells (Supplementary Fig. 18). As evidenced by the patch-clamp measurement, nanocrystalline nanowire exhibits a pronounced photothermal effect, which results in a transient and localized temperature increase of surrounding cytosol and organelles following the laser pulse. This heating effect can either generate reactive oxygen species32 or transiently depolarize/perforate endoplasmic reticulum and mitochondrial membranes33, all of which can trigger the release of calcium from its reservoir to the cytosol. Nevertheless, the same glial cell calcium dynamics can still be modulated repetitively (Supplementary Figs. 19 and 20), indicating the minimal invasiveness of the intracellular stimulation method. Moreover, we observed not only the induced intracellular calcium flux from the glia under direct stimulation, but also the intercellular calcium wave propagation to both neighbouring glia and DRG cells (Fig. 4b). The selective uptake of nanowires by glia and the existence of glia–glia and glia–neuron communication suggest possible remote cellular modulations through naturally occurring intercellular junctions. Finally, the cellular and subcellular dynamics can be exploited for motile modulation biointerfaces (Supplementary Fig. 21).

As Si nanowires can also display active transport along microtubules24, we next explored the possibility of using nanocrystalline Si nanowires as a dual-role intracellular biophysical tool, that is, a calcium modulator and a marker for motor protein–microtubule interactions. We simultaneously tracked the location of a single nanowire (that is, a transport marker) in a glial protrusion and monitored the nearby calcium dynamics, following a remote laser illumination of a different nanowire (that is, a calcium modulator) to initiate a calcium flux within a network (Fig. 4c, first from left). The dynamics of local calcium concentration and the transverse distance of the nanowire, as well as the overlaid time series for both the calcium wave front and the nanowire centre (Fig. 4c, first from right), together suggest a calcium-triggered directional transport of intracellular cargo in the current case. In addition, mean-squared displacement (MSD) analysis (Methods) reveals correlated nanowire transport modes with the local calcium dynamics, that is, from random or restricted diffusions (diffusive exponent, α ≤ 1) without elevated intracellular calcium, to an active transport (diffusive exponent, α ~ 2) after the calcium wave front reached the original nanowire location (Fig. 4c, first from right, and Supplementary Fig. 22). The nanowire transport along the glia protrusion is anterograde, that is, kinesin based. The motor protein kinetics are typically enhanced by increased adenosine triphosphate (ATP) activities34, which may be triggered by the elevation of intracellular calcium concentration35,36.

Besides serving as an intracellular calcium modulator and a transport marker, the photothermal properties of nanocrystalline Si nanowires may be explored to induce a photoacoustic effect for biomechanical manipulation at the subcellular level. To assess this, we chose human umbilical vein endothelial cells (HUVECs), which are active in the phagocytosis of silicon nanowires24 and have well-studied microtubule networks. Nanocrystalline Si nanowires are trapped in the microtubule meshes after co-culturing with HUVECs for ~24 h (Supplementary Fig. 23). When a laser pulse (592 nm, 1 ms, ~2.09 mW, ~211 nm spot size) was introduced to the nanowire, the surrounding microtubules were rapidly repelled and formed a void space near the nanowire (Fig. 4d and Supplementary Fig. 23), suggesting a shock-wave generation through a photoacoustic effect37. Besides intracellular microtubule networks, Si nanowires can also interface with intercellular conduits, where microtubules form compact bundles. On laser illumination of the entangled single nanowire (592 nm, 1 ms, ~2.55 mW, ~211 nm spot size), the bundled microtubules are broken up immediately (Fig. 4e), possibly through a shock-wave-mediated, mechanically induced microtubule depolymerization38. The optically triggered, and nanowire-enabled mechanical manipulation of cytoskeletal structures may serve as a new tool for the study of intra- and intercellular dynamics where a remote structural manipulation of subcellular structures is desired.

Control experiments without nanowires did not yield any of these intra- or intercellular observations. Moreover, the importance of using silicon nanowires instead of other nanostructures (for example, Au nanoparticles or nanorods) is due to the following: (1) silicon nanowires can be at least partially exposed in cytosol on phagocytic cellular entrance24, (2) silicon has only a moderate photothermal effect (compared with, for example, that of Au; Fig. 2b,c and Supplementary Fig. 14) such that the confocal imaging light source itself will not cause heating from the nanostructures and (3) the high aspect ratio of silicon nanowires enables their axial alignment with respect to the cytoskeletal filaments (Fig. 4d,e and Supplementary Fig. 23).

## Single-cell-level and small-tissue-level biointerfaces

In addition to the nanowire-enabled intracellular biointerfaces, we also explored the possibility of implementing extracellular modulations with larger Si structures to match the sizes of cultured cellular assemblies and even small tissues. We first tested p–i–n Si diode junctions because their significantly larger light-induced photocurrents (Figs. 2c,d and 3) may be readily sensed by cells that are attached directly39,40,41. Patch-clamp and calcium imaging studies show that DRGs cultured on Si diode junctions can be stimulated with focused light pulses (Supplementary Figs. 1 a,b and 24) individually or sequentially in a cellular assembly, with a high spatiotemporal resolution and minimal invasiveness (Supplementary Figs. 2527). A pre-immersion of Si substrate in buffer solution for two weeks yielded a twofold increase in threshold laser intensity (Supplementary Fig. 28).

We next explored an in vitro test on acute ex vivo brain slices from mouse neocortex interfacing with a distributed p–i–n Si mesh (Fig. 1e) to evaluate the feasibility of the optically controlled neuromodulation of a small tissue (Fig. 5a, left, and Supplementary Fig. 8c). A whole-cell recording in voltage-clamp mode was made in a cortical pyramidal neuron located in the middle of the 300-μm-thick slice while the Si mesh was in contact with the bottom face of the slice (Fig. 5a, lower right). Immediately after flashing a focused laser beam on the Si mesh (473 nm, 1 ms, ~2 mW, ~57 μm spot size; Fig. 5a, upper right), there were two fast electrical artefacts with opposite polarities (Fig. 5b, marked by #), likely due to the capacitive charging and discharging of the Si/electrolyte/cell interfaces (Fig. 2). Excitatory postsynaptic currents (EPSCs; Fig. 5b, marked by stars) were then recorded arriving with short latency and low jitter after the photoelectric artefacts. Laser illumination of the Si mesh evoked spikes not in the patched neuron but in one or more presynaptic neurons in the slice, which provided the excitatory synaptic input to the recorded postsynaptic neuron. That the patched cell is not triggered to spike may be due to a combined reason that the cell is far away from the Si under illumination (difference in depth of ~150 μm) and only the immediately neighbouring cells may be substantially activated by the localized laser stimulation. The ability to photoactivate presynaptic neurons and detect synaptic inputs with little or no direct activation of the recorded postsynaptic neuron is advantageous for photostimulation mapping of neuronal circuits42,43. The Si mesh in conjunction with focused laser scans thus suggests the potential of this new methodology for ex vivo analysis of brain circuit organization.

## Organ-level biointerfaces

Finally, we interfaced an Au-decorated and Si mesh-based flexible membrane with a mouse brain to control the brain activities, for example, the ability to sense, interpret and act on the environment. The bilayer device layout, consisting of the Au-decorated Si mesh and the holey PDMS membrane (Fig. 1e), allows the device’s conformal attachment to the brain cortex (Supplementary Fig. 29) with sufficient adhesion (Supplementary Fig. 30). We chose an Au-decorated surface due to its large capacitive and Faradic current components. We performed the in vivo photostimulation experiment using an extracellular linear array to record neural activities following laser illuminations (473 nm, ~5 mW, 100 ms, ~216 μm spot size) of the Si mesh attached to the somatosensory cortex of an intact mouse brain (Fig. 5c and Supplementary Figs. 8d and 31). In individual trials of the test, enhanced neural activities were evident during the illumination period—with significant photoelectric artefacts at the light onsets and offsets (Fig. 5d). The detected spike-like events using criteria of a high-pass filter of 800 Hz and a threshold of 5 times the noise level standard deviation exhibit waveforms typical of natural extracellular electrophysiological recordings (Fig. 5e)44,45,46. Peristimulus time histograms (PSTH) from 16 channels (Fig. 5f; with the depths between 200 and 900 μm below the pia) clearly show the illumination-triggered neural responses, in the upper and middle layers of the sensorimotor cortex47. Statistical analyses further revealed that the evoked responses have a significantly higher rate than the spontaneous ones (Supplementary Fig. 31). In addition, stronger short-latency activity was observed in more superficial neurons, which were closer to the Si mesh and thus may be more easily activated. Over time, activity spreads to deeper layers (Fig. 5f), consistent with the propagation of signals through the local and long-range cortical circuits, similar to patterns observed with optogenetic photostimulation48. Moreover, parametric stimulations show a colligative behaviour in that the activated neural response rate is correlated with the stimulation power (Fig. 5g and Supplementary Fig. 32), which is essential to the predictive control of the Si mesh as a precise neuromodulator. Finally, based on the electrophysiology studies, we tested whether the Si mesh-enabled photostimulations of the brain cortex can trigger movements of anaesthetised mice. When we illuminated a Si mesh attached to the right side of the forelimb primary motor cortex, the contralateral left forelimb of the mouse showed a large, rapid up-and-down (flexion-extension) movement shortly after the stimulation (Fig. 5h,i, Supplementary Fig. 33 and Supplementary Videos 1 and 2). Conversely, photostimulation of the left forelimb motor cortex evoked movements of the contralateral right forelimb (Supplementary Fig. 34 and Supplementary Videos 3 and 4). In some cases, small ipsilateral forelimb movements were also evoked (Fig. 5h,i), possibly reflecting activation of uncrossed (ipsilateral) corticospinal projections and/or interhemispheric (callosal) circuits. Overall, our observations of cortically evoked movements are consistent with the functional organization of the forelimb motor control system49 (Supplementary Fig. 35). Given Si membranes yield minimal photothermal effect (Figs. 2 and 3), we believe the combined light-induced capacitive and Faradaic effects from Si induce the observed brain activities and the animal behaviours.

## Discussion

Si-based materials and devices are uniquely suited for light-controlled multiscale biointerfaces (Supplementary Table 1). First, as a semiconductor, Si can induce both the photothermal and photoelectric (including capacitive and Faradaic) processes, whereas metals such as Au display primarily the photothermal effect when plasmonic heating becomes effective (which limits its potential applications for certain biophysical and biomedical studies). Similarly, although carbon and Pt-based materials have been the major systems in electrochemical and related biomedical research, they typically do not display prominent or controllable photoelectric properties, which makes them challenging for the photoelectric biointerfaces shown in this study. Second, both the industrial and synthetic processes for Si are well established, which enables the facile fabrication of various Si forms across multiple length scales. For other semiconductors, such as InP and GaP, although they can also exist in forms similar to Si, their fabrication processes typically require higher costs and more sophisticated instrumentation. In addition, for light-controlled semiconductor-based biointerfaces, the material bandgap is critical as it determines the wavelength range at which the device can be operated. Unlike Si, which absorbs light up to the near infrared region, other common semiconductors, for example, TiO2, GaP, ZnO and CdS, have larger bandgaps and therefore can only work in the ultraviolet–blue region. Finally, Si is not only biocompatible but also biodegradable, which may be important for the future implementation of transient biointerfaces.

We have shown that Si-based materials can behave as optically controlled freestanding devices that modulate brain activities and simple animal behaviours. These are promising complements to two other major neuromodulation techniques, that is, electrical stimulation and optogenetics. We first argue that traditional electrode-based neuromodulation suffers from a few drawbacks that may be addressed by photostimulation. For electrical stimulation, external wirings are required to deliver electrical signals, which would cause additional inflammatory responses in the targeted tissue. Furthermore, the location of the device is fixed by its implantation site and therefore the device, even with advanced materials and layout designs, cannot function in an adjustable manner for the high-resolution stimulation of arbitrary regions. Photostimulation, in contrast, does not require excessive wirings and can be operated with higher flexibility and spatial resolution to implement multiplexed and patterned stimulations on a large scale (that is, the location of stimulation is fully determined by the location of light spots, which can be arbitrarily aimed). Optogenetics, although having the benefit of photostimulation and excellent cellular targeting, requires genetic alteration of the targeted cells or tissue. Existing genetic engineering toolboxes are mostly developed for small animals like rodents, therefore large animals such as non-human primates require additional technical advances in optogenetics50,51,52. In addition, ethical issues of altering human genomes would need to be addressed before optogenetics could be adopted in clinical trials. As an alternative, the Si-based neuromodulation approach combines certain advantages of both electrical stimulation (that is, non-genetic) and optogenetics (that is, optically controlled), offering new opportunities to modulate neural activities in non-human primates or even human subjects.

Looking forward, our current system represents only one step towards remotely controlled non-genetic biological modulations. The future development of other material systems that can effectively transduce external physical inputs (for example, optical, electrical, magnetic and ultrasonic) into output signals (for example, electrical, thermal and mechanical) that are recognizable by the biological systems—in combination with advanced imaging and recording techniques—may ultimately lead to an integrated system for multimodal biological modulations53.

## Methods

### Synthesis of silicon-based materials

Silicon (Si) materials (p-type–intrinsic–n-type diode junctions and intrinsic–intrinsic coaxial nanowires) were prepared using a CVD method. In a typical synthesis of a p–i–n diode junction, a SOI wafer (Ultrasil, device layer, p-type, (100), 0.001–0.005 Ω cm, 2 μm; buried oxide layer, 1 μm; handle layer, p-type, (100), 1–20 Ω cm, 650 μm) was used as the substrate for subsequent deposition of intrinsic and n-type layers. The native oxide on the SOI wafer was removed with hydrofluoric acid (HF, 49%, Sigma-Aldrich) right before placing the substrate inside a quartz tube for evacuation. Each of the intrinsic and n-type Si layers was deposited under 650 °C and a chamber pressure of 15 torr for 20 min. During the intrinsic layer deposition, the flow rates of hydrogen (H2) and silane (SiH4) were set as 60 and 0.3 standard cubic centimetres per minute (sccm), respectively. The n-type layer was deposited with the same flow rates of H2 and SiH4 during the intrinsic layer growth plus a 1.5 sccm flow rate of the dopant gas, phosphine (PH3, 1,000 ppm in H2). Measured conductivities of individual layers are ~0.45 Ω cm for the p-type single-crystalline substrate (~2 µm in thickness), ~149 kΩ cm for the intrinsic polycrystalline layer (~140 nm in thickness), and ~540 Ω cm for the n-type polycrystalline layer (~190 nm in thickness), respectively.

The nanocrystalline Si nanowire (intrinsic core/intrinsic shell) was synthesized from a two-step process involving an initial growth of a thin intrinsic backbone and a subsequent deposition of a thick intrinsic shell. The core was grown with a gold (Au) nanocluster-catalysed CVD process where Au colloidal nanoparticles (Ted Pella, 50 nm in diameter) were deposited onto a Si (100) substrate (Nova Electronic Materials, n-type, 0.001–0.005 Ω cm) as the catalyst. The growth was maintained at 470 °C and 40 torr for 20 min. The flow rates of H2 and SiH4 were controlled as 60 and 2 sccm, respectively. After the intrinsic nanowire core growth, the SiH4 flow was switched off and the chamber was kept under a H2 atmosphere (60 sccm, 15 torr) until the temperature ramped up to 600 °C for the subsequent shell deposition. The H2 atmosphere was used to minimize Au diffusion. The intrinsic shell was deposited with flow rates of H2 and SiH4 at 0.3 and 60 sccm, respectively, and a chamber pressure of 15 torr for 40 min.

The metal-decorated Si diode junctions were prepared with an electroless deposition method. In general, the as-synthesized p–i–n diode junctions were dipped into a mixture of metal-containing solutions (chloroauric acid (HAuCl4), potassium tetrachloroplatinate(ii) (K2PtCl4), silver nitrate (AgNO3); 0.01 mM, 0.1 mM, 1 mM) and 1% HF for 3 min at room temperature.

### Fabrication of Si-based flexible devices for in vivo experiments

The device fabrication process was divided into two parallel steps including the preparations of both distributed Si meshes and porous PDMS (Corning) membranes (Supplementary Fig. 6). The fabrication of distributed Si meshes were performed with a combination of photolithography and etching techniques. In brief, a bilayer of undercut (MicroChem, LOR-3A) and photoresist (MicroChem, SU-8 2005) was spin-coated on the as-synthesized p–i–n diode junction SOI wafer. A mesh structure of SU-8 was patterned with a standard photolithography process consisting of ultraviolet light exposure (200 mJ cm−2) and developing (MicroChem, SU-8 developer). The as-patterned SU-8 mesh served as an etch mask for the subsequent reactive ion etching of Si. The unprotected p–i–n Si layers (~2.3 μm) were removed after 10 min of etching (radiofrequency power, 100 W; inductive-coupled plasma power, 400 W) with a gaseous mixture of tetrafluoromethane (CF4, 45 sccm) and argon (Ar, 5 sccm). The SU-8 protection layer was lift-off by dissolving the undercut LOR-3A layer in Remover-PG (MicroChem). A final wet etching of the oxide layer with 49% HF was performed to release the as-patterned Si diode junction.

The PDMS membrane was prepared using a soft-lithography technique. In general, a SU-8 pillar array (~120 μm in height) was patterned on a Si substrate with the standard photolithography process and served as the soft-lithography mould. A layer of PDMS (precursor:curing agent ratio = 10:1) was then spin-coated onto the SU-8 mould and cured at 80 °C overnight. The as-cased PDMS layer was finally released in hexane (Fisher Scientific) to get the holey structure. The distributed Si mesh was then transferred onto the holey PDMS membrane to form the entire device (Supplementary Fig. 6).

### Finite element analysis of stress distribution

The finite element analysis was performed using COMSOL Multiphysics 5.3 (COMSOL, https://www.comsol.com). A two-dimensional plate model in the structural mechanics module was set for the simulation. The von Mises stress distribution was calculated after a point displacement of 500 μm in the z direction at the device centre while fixing four edge points.

### Electron microscopy

A transmission electron microscope (JEOL, JEM-3010) and an aberration-corrected scanning transmission electron microscope (JEOL, JEM-ARM200F) were used to image the cross-sectional structures of both the pristine p–i–n Si diode junction and the gold-decorated one. X-ray energy dispersive spectroscopy maps were using the JEM-ARM200F, which was equipped with an Oxford X-MaxN 100TLE windowless SDD X-ray detector (Oxford Instruments). Selected area electron diffraction (SAED) patterns were taken using the JEM-3010. TEM cross-sectional specimen preparations were carried out by controlled tripod polishing54,55 followed by liquid-nitrogen-cooled Ar ion millings using a Fischione 1050 TEM mill (Fischione Instruments). A 4 kV ion milling was used to further thin the specimen and a final 0.5 kV milling was performed to remove surface damages. A scanning electron microscope (Carl Zeiss, Merlin) was used to image the top view of the Au-decorated Si diode junction, the cross-sectional view of the diode junction, and the flexible device made of the distributed Si mesh and the holey PDMS membrane. Nanocrystalline Si nanowires were sonicated in isopropanol (Sigma-Aldrich) and then dispersed over copper grids (Ted Pella, Lacey Formvar/Carbon, 200 mesh) for side-view imaging using a transmission electron microscope (JEOL, JEM-3010). The cross-sections of the nanowires were prepared by ultramicrotomy. In general, Si nanowires were embedded in epoxy resins which were then solidified at 60 °C for 24 h. Thin epoxy sections of ~100 nm were cut using a ultramicrotome (Ultracut E, Reichert-Jung), collected on lacey carbon grids (Ted Pella), and imaged using the same transmission electron microscope.

### X-ray photoelectron spectroscopy

X-ray photoelectron spectroscopy data were collected using ESCALAB 250 Xi (Thermo Scientific) with a monochromatic Al Kα ( = 1,486.6 eV) excitation. The diameter of the X-ray beam was 500 µm. The survey scans were performed with a pass energy of 160 and a step size of 1 eV, whereas the high-resolution scans were done with a pass energy of 50 and step size of 0.1 eV. The correction of the X-ray photoelectron spectra for charge accumulation was performed using the Si 2p peak (binding energy = 99.4 eV). The Si 2p peaks were fitted using a Shirley background with G/L 30% for Si0 and pure Gaussian for Si4+. Peak fitting for the Au 4f signal was determined with a linear background G/L 30%, asymmetric 0.9, and a height factor of 0.75. With these parameters the full width at half maximum range for the gold was from 0.8 eV to 1.25 eV.

### Si photoresponse measurements

For the photoresponse measurements, a standard patch-clamp setup was employed. In particular, an upright microscope (Olympus, BX61WI) with a 20×/0.5 numerical aperture (NA) water-immersion objective was used to deliver light pulses from a LED (M530L3, Thorlabs, 530 nm, ~500 μm spot size) or a laser (Laserglow, 532 nm, diode-pumped solid-state laser, ~5 μm spot size). The light pulses were controlled by transistor–transistor logic signals (10 ms) delivered from a digitizer (Molecular Devices, Digidata 1550). Voltage-clamp protocols were done by an Axopatch 200B amplifier (Molecular Devices), controlled by pClamp software (Molecular Devices). Glass pipettes were pulled in a flaming/brown type micropipette puller (Sutter Instrument, P-97) for a final resistance of ~1 MΩ when filled with 1 × PBS (Fisher Scientific) solution. In a typical measurement, a Si material was immersed in the same PBS solution where the pipette tip was positioned in close proximity to the Si surface (~2 μm). The ionic currents across the pipette tip were recorded in the voltage-clamp mode where the holding levels of the pipette were adjusted using the pipette offset knob. The individual quantities of each photoresponse, that is, capacitive, Faradaic and thermal, were calculated by fitting the plot of the light-induced current amplitude (ΔIlight) over the holding level (I0). At a given time point, the slope of the ΔIlight − I0 plot represents the photothermal response whereas the intercept of the plot is contributed by the photoelectric responses. Within the photoelectric responses, the capacitive current is defined as the maximal current amplitude after the light onset while the Faradaic current is defined as the current amplitude at the time point of 8.5 ms since illumination starts. The amplitude of the photothermal-induced local temperature increase of the solution was calculated after the calibration of the pipette resistance. After the photoresponse measurement, the same micropipette was placed into another dish of pre-heated PBS with an initial temperature of about 50 °C. A thermocouple was positioned close to the pipette tip during the temperature measurement. A calibration curve was created, based on the pipette resistance changes in the range between 50 and 20 °C, which was then used in conjunction with the ΔIlight − I0 curve to estimate the local temperature increase.

### Mammalian cell cultures

All animal protocols used were in accordance with the policies of the University of Chicago and were approved by the Institutional Animal Care and Use Committees (IACUC).

#### DRG culture

DRGs were extracted from decapitated P1-P3 Sprague-Dawley rats (Charles River Laboratories) and were placed immediately in ice-cold Dulbecco’s modified eagle medium (DMEM/F12, Life Technologies). The ganglia were then transferred to a 2.5 mg ml−1 trypsin solution (Worthington) in Earle's balanced salt solution (EBSS, Life Technologies) and digested for 20 min in a 37 °C shaker with a speed of 144 r.p.m. Afterwards, the cells were centrifuged and the supernatant was replaced with EBSS supplemented with 10% fetal bovine serum (FBS; ATCC). After the mechanical trituration with pipetting, the cell suspension was centrifuged again and the supernatant was replaced with DMEM/F12 containing 5% FBS. Next, cells were seeded onto poly-l-lysine (PLL, Sigma-Aldrich)-coated substrates, for example, glass-bottom Petri dishes, p–i–n diode junction SOI wafers, and allowed 30 min for cell adhesion. Finally, the dishes were filled with DMEM/F12 supplemented with 5% FBS, 100 U ml−1 penicillin (Sigma-Aldrich), and 100 μg ml−1 streptomycin (Sigma-Aldrich), and cultured in a 37 °C incubator with 5% carbon dioxide (CO2) until used for experiments.

#### Other cell lines

HUVEC (Life Technologies) and U2OS (ATCC) cells were cultured on glass-bottomed Petri dishes and passaged following standard procedures from the vendors.

#### Cellular internalization of Si nanowires

In all cell cultures, intrinsic nanocrystalline Si nanowires were introduced and allowed to co-culture for at least 24 h. Specifically, nanowire suspensions in different cell culture media were prepared by extensively sonicating small pieces of nanowire growth substrates (~2 mm × 2 mm) in culture media for 2 min. The as-made nanowire suspensions were added to the cultures in a drug-like fashion (~10 μl of suspension per 1 ml of medium). Before all experiments, cells were washed three times with fresh media.

### Immunofluorescence labelling of the DRG culture

DRG and nanowire cocultures were first fixed with 4% paraformaldehyde in PBS (Alfa Aesar, with magnesium and ethylene glycol tetraacetic acid) for 10 min at room temperature. After rinsing in PBS, cells were then permeabilized with 0.1% Triton X-100 in PBS (Sigma-Aldrich) for another 10 min at room temperature. Following blocking with 1.5% bovine serum albumin (BSA, Sigma-Aldrich) in PBS for 1 h, the cells were incubated with primary antibodies (GFAP (GA5) Mouse mAb, 1:300 in 1.5% BSA-PBS for glia; NeuN (D4G40) XP Rabbit mAb, 1:50 in 1.5% BSA-PBS for neuron, Cell Signaling) at room temperature for 1 h. After washing, secondary antibodies (Goat anti-Mouse IgG (H + L) Superclonal Secondary Antibody, Alexa Fluor 647, 1:150 in 1.5% BSA-PBS for glia; Goat anti-Rabbit IgG (H + L) Secondary Antibody, Alexa Fluor 488, 1:150 in 1.5% BSA-PBS for neuron, Life Technologies) were finally applied. As NeuN is expressed in the neuronal nucleus, another set of biomarkers, that is, S-100 for glial cells and Neurofilament for neurons, were stained to test whether nanowires were colocalized with neuronal cytoplasm. The staining follows the same procedure as the GFAP/NeuN staining with slight differences in the dilution ratios of the antibodies. In particular, cells after fixation and permeabilization were incubated with primary antibodies (S100 Polyclonal Antibody, 1:100 in 1.5% BSA-PBS for glia, Life Technologies; Neurofilament-H (RMdO 20) Mouse mAb, 1:200 in 1.5% BSA-PBS for neuron, Cell Signaling) at room temperature for 1 h. After washing, secondary antibodies (Goat anti-Mouse IgG (H + L) Cross-Adsorbed Secondary Antibody, Alexa Fluor 647, 1:200 in 1.5% BSA-PBS for neuron; Goat anti-Rabbit IgG (H + L) Cross-Adsorbed Secondary Antibody, Alexa Fluor 488, 1:200 in 1.5% BSA-PBS for glia, Life Technologies) were finally applied. After washing in PBS, the as-labelled cells were imaged using a confocal laser scanning microscope (Leica, SP5 II STED-CW) with Si nanowires being imaged simultaneously with the scattered light. The staining results using GFAP and NeuN as the markers were included in Fig. 4a and Supplementary Fig. 15 while the staining results using S-100 and Neurofilament were included in Supplementary Fig. 16.

### Calcium imaging

Cells, either co-cultured with intrinsic nanowires or cultured on p–i–n diode junctions, were stained with 2 μM of Fluo-4 AM (Life Technologies) for 30 min at 37 °C and washed three times with dye-free culture media before imaging. The as-stained cells were imaged using the same Leica SP5 confocal microscope. In a typical experiment, a laser pulse (1 ms, 592 nm) was delivered to the nanowire/cell of interest in the middle of a calcium imaging time series. The cellular fluorescence intensity over time was then analysed using ImageJ software (National Institutes of Health, https://imagej.nih.gov/ij/).

Cells, either co-cultured with intrinsic nanowires or cultured on p–i–n diode junctions, were stained with 2 μM of calcein AM (Life Technologies) and 4 μM of ethidium homodimer-1 (Life Technologies) for 30 min at room temperature. The as-stained cells were imaged using the same Leica SP5 confocal microscope. In a typical experiment, a laser pulse (1 ms, 592 nm) was delivered to the nanowire/cell of interest in the middle of an imaging time series. Live cells will stay green throughout the entire sequence while dead cells will be stained red.

### Intracellular transport

For the intracellular transport study, a custom-written programme in Python was run to automatically track the nanowire and calcium wave-front locations in all 100 frames. To distinguish between different modes of Si nanowire transport, we used a rolling frame MSD metric, where the MSD is the average distance that a particle travels as a function of lag time, given by:

$${\mathrm{MSD}} = \left\langle {\Delta r^2(\tau )} \right\rangle = q\tau ^\alpha$$

where Δr, τ, q and α are the nanowire displacement, lag time, diffusion coefficient and the ‘diffusive exponent’, respectively. The diffusive exponent, α, can be used as a metric of transport properties, distinguishing between Brownian diffusion (α = 1), restricted diffusion (α < 1) and active transport (α > 1) processes. For plotting, rolling α values were used to yield an estimate of the local transport behaviour.

### Live cell microtubule dynamics

In a typical experiment, HUVEC cells with internalized Si nanowires were stained with 200 nM of SiR-tubulin (Cytoskeleton) at 37 °C for 1 h. Verapamil (10 μM; cytoskeleton) was also added to inhibit the efflux of the SiR-tubulin. Three times of washing with the dye-free medium was applied before imaging. Under the same Leica SP5 confocal microscope, the Si nanowire of interest was illuminated with a 592 nm laser pulse (1 ms) and the subsequent microtubule dynamics were recorded. The as-recorded videos were processed and analysed using ImageJ including the analysis of microtubule bounded areas and intercellular conduit lengths over time, and the generation of kymographs.

### Electrophysiology and photostimulation experiments

All animal protocols used were in accordance with the policies of the University of Chicago and Northwestern University, approved by the Institutional Animal Care and Use Committees (IACUC), and followed the animal welfare guidelines of the Society for Neuroscience and National Institutes of Health.

#### DRG culture on Si diode junctions

Before the experiment, FBS supplemented DMEM/F12 in the culture dish was rinsed three times with the extracellular recording solution (in mM: NaCl 132, KCl 4, MgCl2 1.2, CaCl2 1.8, HEPES 10, glucose 5.5; pH 7.4). Desired neurons were patched with a ~2 MΩ pipette, filled with the intracellular pipette solution (in mM: NaCl 10, KCl 150, MgCl2 4.5, EGTA 9, HEPES 10; pH 7.3). Voltage recordings were made in current-clamp mode using the same setup for the photoresponse measurements. Suprathreshold current injections were first delivered to the patched neuron to assess its excitability. Laser pulses (532 nm) with incremental durations were delivered subsequently to excite the cell.

#### Brain slice with Si diode junctions

Wild-type mice (C57BL/6, female and male; Jackson Laboratory) were bred in-house. Mice were 6–9 weeks old at the time of the slice experiments.

### Slice preparations

Mice were euthanized by anaesthetic overdose anddecapitation. Brain slices were made in a 4 °C cutting solution (in mM: 110 choline chloride, 11.6 sodium Lascorbate, 3.1 pyruvic acid, 25 NaHCO, 25 d-glucose, 2.5 KCl, 7 MgCl, 0.5 CaCl, 1.25 NaH PO; aeratedwith 95% O/5% CO) using a vibratome (VT 1200S, Leica) to make 250-μm-thick slices. The slices weretransferred to an artificial cerebrospinal fluid (ACSF, composition in mM: 127 NaCl, 25 NaHCO, 25 d-glucose, 2.5 KCl, 1 MgCl, 2 CaCl, 1.25 NaH PO; aerated with 95% O/5% CO) and maintained at 34 °C for 30 min.The slices were then returned to room temperature for at least 1 h before the recordings.

### Electrophysiology

Distributed Si meshes were placed in a recording chamberunder an upright microscope (BX51WI; Olympus), which was equipped with a video camera (Retiga 2000R;QImaging). Brain slices were then transferred on top of the Si meshes to form contacts. Slices werevisualized by bright-field gradient contrast microscopy using an infrared LED (850 nm, M850L2, Thorlabs) asthe light source. A low-magnification objective lens (UPlanSApp 4×/NA 0.16, Olympus) was used to visualizeand position the slices. A high-magnification water-immersion lens (LUMPlanFLN 60×/NA 1.00, Olympus,Japan) was used to identify neurons for whole-cell recordings.

Borosilicate glass (inner diameter 0.86 mm, outer diameter 1.5 mm with filament, Warner Instruments) was pulled using a P-97 micropipette puller (Sutter Instrument) into patch pipettes with a tip resistance of 2–4 MΩ. Neurons targeted for whole-cell recordings were obtained using micromanipulators (MP-225, ROE-200, MPC-200, Sutter Instrument) and a patch-clamp amplifier (Multiclamp 700B, Axon Instruments). Pipettes containing potassium-based or caesium-based internal solutions were used for voltage-clamp recordings (composition of the internal solution, in mM: 128 potassium or caesium methanesulfonate, 10 HEPES, 10 phosphocreatine, 4 MgCl2, 4 ATP, 0.4 GTP, 3 ascorbate, 1 EGTA, 1 QX-314, and 0.05 Alexa Flour hydrazide, with 4 mg ml−1 biocytin, at 7.25 pH and 290–295 mOsm). All recordings were made in 34 °C ACSF with the temperature controlled by an in-line feedback-controlled heater (TC 324B, Warner Instruments). Recordings with series resistance >40 MΩ were excluded.

A command potential of −70 mV was applied to isolate excitatory (glutamatergic) post synaptic currents (EPSCs). To test for input to a neuron, blue-laser illuminations (1 ms long pulses, 473 nm, ~2 mW, ~57 μm spot size; MLL-FN473, CNI Laser) were delivered onto a nearby spot of the Si mesh.

Multiple trials were sampled at an interstimulus interval of at least 30 s. Recorded currents were amplified, filtered at 4 kHz, and sampled at 40 kHz. Data were acquired using Ephus software56 and analysed using routines written in MATLAB (MathWorks).

#### In vivo experiments with distributed Si meshes

Wild-type mice (C57BL/6, female and male; Jackson Laboratory) were used, at an age of 6–9 weeks old at the time of the in vivo experiments.

### Pre-stimulation surgeries

The mouse was deeply anaesthetised with ketamine-xylazine (ketamine 80–100 mg  kg−1, xylazine 5–15  mg  kg−1, injected intraperitoneally) before the placement of the cranial mounting hardware. A small skin incision was first made over the cerebellum to expose the skull. A stainless-steel set screw (single-ended #8-32, SS8S050, Thorlabs), crimped with a spade terminal (non-insulated, 69145K438, McMaster-Carr) was then affixed with dental cement to the skull. This set screw was later screwed into a tapped hole located at the top of a 1/2ʺ optical post for the head fixation.

After being head-fixed as described above, craniotomies were made over the motor and somatosensory cortices using a dental drill with large enough openings (~2.5   mm) to allow the attachment of a Si mesh on the cortex and the passage of a linear probe. The dura was peeled for a full exposure of the cortex, which was important for a good signal transduction at the Si/brain interface. The mouse was then placed in the recording apparatus with the body temperature monitored with a rectal probe and maintained at ~37.0   °C via a feedback-controlled heating pad (FHC, Bowdoin). During the subsequent recordings, ACSF was frequently applied to the exposed brain area to prevent the damage from dehydration. The level of anaesthesia was continuously monitored based on whisker movements and paw-pinching/eye-blinking reflexes. Additional anaesthetics with 50% of the induction dosage were given when required.

### Photostimulation apparatus

A customized laser scanning apparatus with a blue-laser source (LY473III-100, wavelength 473  nm, maximum power ~100  mW, beam diameter ~2  mm) mounted on a three-dimensional linear stage was positioned above the mouse head. In the apparatus, the laser beam from the light source goes through an acousto-optic modulator (AOM) and an iris before being deflected by a pair of galvanometer scanners and focused to the Si mesh by a plano-convex spherical lens.

The output laser power was controlled using a customized AOM driver modulated by signal waveforms delivered via a commercial multifunction (analogue and digital) interface board (NI USB 6229, National Instruments). A short pulse train was also sent to digitally encode the parameters of the light waveform such as the start point through the digital input port of the electrophysiology data acquisition (DAQ) board. Software tools (LabVIEW, National Instruments) including a graphical user interface (GUI, GenWave) were developed to generate and transfer waveforms to the AOM driver. The system was calibrated using a power metrer to determine the relationship between the driver input voltage and the laser scanner output power.

### Electrophysiology apparatus

Silicon probes of 32-channel linear microelectrodearrays with ~1 MΩ impedances and 50  µm spacings (model A1×32-6mm-50-177, NeuroNexus) were used for electrophysiological recordings. The probe was fixed to a motorized four-axis micromanipulator, assembled by mounting a MTSA1 linear translator (Thorlabs) onto a MP285 3-axis manipulator (Sutter Instrument), and positioned under stereoscopic visualizations over a distributed Si mesh that has been attached to the cortical surface (with the Si layer facing towards the tissue). The probe was tilted by ~30° off the vertical axis for a better collection of the neural signals under the silicon mesh. The probe was then slowly inserted into the cortex at a rate of 2  µm  s−1 controlled by LabVIEW, until it reached a depth of 1,600 µm from the pia, with the entry point in the sensorimotor cortex adjacent to the edge of the silicon mesh. Laser pulses with various powers (up to 5  mW, ~216  μm spot size) and durations (up to 100 ms) were delivered onto the Si mesh for the photostimulation of the brain.

Signals were amplified using a RHD2132 amplifier board based on a RHD2132 digital electrophysiology interface chip (Intan Technologies). The filter was set to an analogue bandpass of 0.1–7.5 kHz with a digital filter cutoff of 1 Hz. The single channel sample rate was set to 30 K SPS.

For hardware control, we used a RHD2000 USB Interface Board (Intan Technologies) for the communication with other digital devices and the streaming of all the neural-signal data from the RHD2000 amplifiers. The USB port of the module was linked with a USB cable to pipe the data stream in to and out of the computer. In this experiment, the digital ports included in the DAQ board were only used for the acquisition of the photostimulation parameters from the AOM controller.

C++/Qt based experimental interface software (Intan Technologies) was used for the amplifier configuration, online visualization and data logging.

### Forelimb movement study apparatus

A Chameleon3 USB3 CMOS Mono camera (CM3-U3-13Y3M-CS, FLIR Systems) configured at 640  ×  512 pixels (2  ×  2 binning) was used to record the body movements following the laser stimulations. The video recording was triggered and synchronized by the laser scanning control board with the frame rate of 100  Hz. Fifty frames were collected before the start of the stimulation and a total of 100 frames were recorded for a full trial. A fixed focal length lens (35  mm EFL, f/2.0, Navitar) was mounted on the camera for the focusing. The centroids of the mouse claws were tracked in each frame to investigate the forelimb movements following the laser stimulations. The trajectories of the centroids were quantified to illustrate the movements in each trial.

### Data analyses

The recorded data were stored as raw signals from the amplifiers and filtered by a 60  Hz notch filter. To reduce the contaminations of the probe recording signals due to the strong photovoltaic effect of the Si mesh, we used a digital high-pass filter (800  Hz cutoff, second-order Butterworth), to shrink the photovoltaic artefact to the first 3  ms post-stimulus window.

The following routines were performed to further analyse the data. First, a threshold detector was applied, with the threshold set to the five times the standard deviation to detect the spikes. To mask the photovoltaic effect, spike counts of the first 3 ms window were then replaced by null values. Finally, neural response time stamps were determined for each detected spike and the response waveforms were plotted from −0.67 to 1.33 ms with respect to the detected spike time stamp, that is, 20 points before and 40 points after the spike time stamp with a sampling rate of 30 kHz. The detected waveforms were sorted according to the similarity of the shapes, that is, peak to valley amplitudes of the responses. All the analysis codes were written in Matlab (Mathworks).

The time stamps of all the spikes from each channel were used to generate the peristimulus time histogram and the heat maps, which represent the instantaneous firing rate, with 1 ms binning. Responses were averaged across all trials in each channel to yield a mean histogram.

### Micro computed tomography of the Si/brain interface

Micro computed tomography images of gold-decorated Si meshes attached to dead mouse brains were performed on the XCUBE (Molecubes NV.) by the Integrated Small Animal Imaging Research Resource (iSAIRR) at the University of Chicago. Images were acquired with an X-ray source of 50 kVp and 200 µA in a single frame of 960 projections. Volumetric CT images were reconstructed in a 400 × 400 × 400 format with voxel dimensions of 100 µm3. Images were analysed using AMIRA 5.6 (Thermo Fisher Scientific).

Adult C57BL/6 mice (Jackson Laboratory) were killed shortly before the mechanical test. Mouse brains were collected from dead animals and placed inside PBS solutions before the adhesion tests by a tensile test machine (Zwick-Roell, zwickiLine Z0.5). Briefly, the brain was fixed on a glass slide using a tissue adhesive (Ted Pella, Pelco Pro CA44) and the device was held tightly by a grip. After forming a conformal contact between the device and the brain cortex with an area of ~8 mm × 4 mm, a unidirectional tension was applied to peel the device off the brain while the force and the extension were recorded simultaneously. The loading rate was kept constant at 3 mm min−1. The adhesion energy per area was calculated by the integration of the force–extension curves divided by the contact areas.

### Analysis of the photoresponse measurements

In a typical photoresponse measurement, a glass micropipette, with the potential holding at a fixed level (Vp), was positioned near a Si material surface immersed in PBS, and a 10-ms-long light pulse was delivered to the material in the middle of the trial. Therefore, the recorded current across the pipette tip can be divided into two parts, namely the baseline current I0 at the dark stage and the time-dependent light-generated current ΔIlight(t) at the light stage. During the light illumination period, two parallel processes originated from the Si material can contribute to ΔIlight(t).

The first is the photoelectric process where the light-generated excessive carriers will accumulate on the Si surface and change the local surface potential. Ions in the nearby medium will be attracted/repelled with respect to this photogenerated potential and create the ionic currents. As the variation of the surface potential is only a function of the carrier dynamics on the Si surface, the photoelectrically induced ionic current (ΔIelectric(t)) is therefore independent of the holding current level I0.

Another process is related to the photothermal effect of Si where the recombination of light-generated carriers converts part of the input photon energy into the vibrational energy of the Si lattice, which dissipates heat through both Si and the surrounding electrolyte. For the electrolyte with an increased temperature, mobilities of the ions will increase, resulting in a reduced pipette tip resistance R. Even under a fixed holding potential Vp, the current during the light illumination period will change due to the decrease of the pipette resistance. Therefore, the thermally induced current (ΔIthermal(t)), is strongly related to the holding potential Vp and the baseline current I0, where $$V_{\mathrm{p}} = I_0 \times R_0$$, $$R_0$$ is the pipette resistance in dark.

Given the significantly different dependences on the holding current I0 for the electrically (ΔIelectric(t)) and thermally induced currents (ΔIthermal(t)), the total light-generated currents (ΔIlight(t) = ΔIelectric(t) + ΔIthermal(t)) can be potentially decoupled by analysing current traces recorded at different holding levels.

At a given time point t during the light illumination period, the recorded current, I0 + ΔIlight(t), excluding the photoelectrically induced current part ΔIelectric(t), and the pipette tip resistance R(t) follow the Ohm’s law as long as the holding potential Vp is fixed.

$$V_{\mathrm{p}} = I_0 \times R_0 = \left( {I_0{\mathrm{ + }}\Delta I_{{\mathrm{light}}}(t) - \Delta I_{{\mathrm{electric}}}(t)} \right) \times R(t)$$
(1)

Rearranging equation (1) gives the relationship between the light-induced current ΔIlight(t) and the holding current I0 that:

$$\Delta I_{{\mathrm{light}}}(t) = \left( {\frac{{R_0}}{{R(t)}} - 1} \right) \times I_0 + \Delta I_{{\mathrm{electric}}}(t)$$
(2)

As shown in equation (2), the photoelectric effect is explicitly manifested as the intercept of the curve. A photocurrent plot can be generated by plotting the fitted intercept values over time as shown in Fig. 2 and Supplementary Figs. 9, 11 and 14. Two types of photoelectric responses, that is, capacitive and Faradaic, are further identified based on the dynamics and the amplitude of the currents. Two spiky features at the onset and offset of the light illumination are capacitive currents corresponding to the capacitive charging/discharging processes at the Si/electrolyte interface. A long-lasting current with a lower amplitude is the Faradaic current due to the surface redox reactions.

The photothermal effect, in contrast, is implicitly embedded in the fitted slope as the pipette resistance is a function of temperature. To calculate the photothermally induced temperature change of the surrounding medium, a calibration curve of the pipette resistance over temperature is needed, which typically follows an Arrhenius-type relationship57 that:

$$\ln {R} = a \times \frac{1}{T} + c$$
(3)

where a and c represent the slope and intercept values.

In conjunction with the slope k(t) from the ΔIlight(t) − I0 plot that:

$$R(t) = \frac{{R_0}}{{k(t) + 1}}$$
(4)

the final temperature of the surround medium heated from the photothermal effect is determined only by the slopes of the ΔIlight(t) − I0 and the lnR − 1/T curves that:

$$T(t) = \frac{1}{{\frac{1}{{T_0}} - \frac{1}{a}\ln (k(t) + 1)}}$$
(5)

Notably, as the photothermal effect is a function of the illumination duration, the slope of the ΔIlight(t) − I0 plot is also time dependent. The maximal temperature is reached after 10 ms of illumination so ΔIlight,10 ms − I0 plots were used to assess the photothermal responses of various Si materials presented in Figs. 2 and 3 and Supplementary Figs. 9, 11, 13 and 14. A temperature over time curve can also be generated using the fitted slope values from each time point as shown in Fig. 2.

In summary, the ΔIlight,10 ms − I0 plot method can be applied to virtually all kinds of materials other than just Si to assess their photoresponses, which will fall into the following four categories.

(1) In one extreme case where the material has only the photothermal effect without any photoelectric effect, that is, ΔIelectric = 0 at all time, equation (2) will be reduced to:

$$\Delta I_{{\mathrm{light}},{\mathrm{10}}\,{\mathrm{ms}}} = \left( \frac{{R_{0}}}{R_{10\,{\mathrm{ms}}}} - 1 \right) \times I_{0}$$
(6)

The ΔIlight,10 ms − I0 plot will be a slanted line with a zero intercept.

(2) In another extreme scenario where the material has only the photoelectric effect without any photothermal effect, that is, R0 = R10 ms, equation (2) can be written as:

$$\Delta I_{{\mathrm{light,10}}\,{\mathrm{ms}}} = \Delta I_{{\mathrm{electric,10}}\,{\mathrm{ms}}}$$
(7)

The ΔIlight,10 ms − I0 plot will be a horizontal line with a non-zero intercept.

(3) If a material does not have any photoresponses, equation (2) will be:

$$\Delta I_{{\mathrm{light,10}}\,{\mathrm{ms}}} = 0$$
(8)

The ΔIlight,10 ms − I0 plot will be a horizontal line with a zero intercept.

(4) In any intermediate situations where both the photoelectric and the photothermal effects coexist, the original form of equation (2) applies that:

$$\Delta I_{{\mathrm{light,10}}\,{\mathrm{ms}}} = \left( {\frac{{R_0}}{{R_{10\,{\mathrm{ms}}}}} - 1} \right) \times I_0 + \Delta I_{{\mathrm{electric,10}}\,{\mathrm{ms}}}$$
(9)

The ΔIlight,10 ms − I0 plot will be a slanted line with a non-zero intercept.

### Reporting Summary

Further information on experimental design is available in the Nature Research Reporting Summary linked to this article.

### Code availability

Custom codes used to analyse the data are available from the corresponding author upon reasonable request.

### Data availability

The authors declare that all data supporting the findings of this study are available within the paper and its Supplementary Information. Other supporting data are available from the corresponding author upon request.

Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

## References

1. 1.

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

2. 2.

Tian, B., Cohen-Karni, T., Qing, Q., Duan, X., Xie, P. & Lieber, C. M. Three-dimensional, flexible nanoscale field-effect transistors as localized bioprobes. Science 329, 830–834 (2010).

3. 3.

Liu, J. et al. Syringe-injectable electronics. Nat. Nanotech. 10, 629–636 (2015).

4. 4.

Chiappini, C. et al. Biodegradable silicon nanoneedles delivering nucleic acids intracellularly induce localized in vivo neovascularization. Nat. Mater. 14, 532–539 (2015).

5. 5.

Tian, B. et al. Macroporous nanowire nanoelectronic scaffolds for synthetic tissues. Nat. Mater. 11, 986–994 (2012).

6. 6.

Fang, H. et al. Capacitively coupled arrays of multiplexed flexible silicon transistors for long-term cardiac electrophysiology. Nat. Biomed. Eng. 1, 0038 (2017).

7. 7.

Colicos, M. A., Collins, B. E., Sailor, M. J. & Goda, Y. Remodeling of synaptic actin induced by photoconductive stimulation. Cell 107, 605–616 (2001).

8. 8.

Tee, B. C. et al. A skin-inspired organic digital mechanoreceptor. Science 350, 313–316 (2015).

9. 9.

Khodagholy, D. et al. NeuroGrid: recording action potentials from the surface of the brain. Nat. Neurosci. 18, 310–315 (2015).

10. 10.

Fu, T. M. et al. Stable long-term chronic brain mapping at the single-neuron level. Nat. Methods 13, 875–882 (2016).

11. 11.

Chortos, A., Liu, J. & Bao, Z. Pursuing prosthetic electronic skin. Nat. Mater. 15, 937–950 (2016).

12. 12.

Someya, T., Bao, Z. & Malliaras, G. G. The rise of plastic bioelectronics. Nature 540, 379–385 (2016).

13. 13.

Chiappini, C. et al. Biodegradable nanoneedles for localized delivery of nanoparticles in vivo: exploring the biointerface. ACS Nano 9, 5500–5509 (2015).

14. 14.

Zimmerman, J. F. et al. Free-standing kinked silicon nanowires for probing inter- and intracellular force dynamics. Nano Lett. 15, 5492–5498 (2015).

15. 15.

Chu, B., Burnett, W., Chung, J. W. & Bao, Z. Bring on the bodyNET. Nature 549, 328–330 (2017).

16. 16.

Sakimoto, K. K., Wong, A. B. & Yang, P. Self-photosensitization of nonphotosynthetic bacteria for solar-to-chemical production. Science 351, 74–77 (2016).

17. 17.

Liu, C., Colon, B. C., Ziesack, M., Silver, P. A. & Nocera, D. G. Water splitting-biosynthetic system with CO2 reduction efficiencies exceeding photosynthesis. Science 352, 1210–1213 (2016).

18. 18.

Chen, R., Romero, G., Christiansen, M. G., Mohr, A. & Anikeeva, P. Wireless magnetothermal deep brain stimulation. Science 347, 1477–1480 (2015).

19. 19.

Seo, D. et al. Wireless recording in the peripheral nervous system with ultrasonic neural dust. Neuron 91, 529–539 (2016).

20. 20.

Nadeau, P. et al. Prolonged energy harvesting for ingestible devices. Nat. Biomed. Eng. 1, 0022 (2017).

21. 21.

Jiang, Y. et al. Heterogeneous silicon mesostructures for lipid-supported bioelectric interfaces. Nat. Mater. 15, 1023–1030 (2016).

22. 22.

Grossman, N. et al. Noninvasive deep brain stimulation via temporally interfering electric fields. Cell 169, 1029–1041 (2017).

23. 23.

Dagdeviren, C. et al. Flexible piezoelectric devices for gastrointestinal motility sensing. Nat. Biomed. Eng. 1, 807–817 (2017).

24. 24.

Zimmerman, J. F. et al. Cellular uptake and dynamics of unlabeled freestanding silicon nanowires. Sci. Adv. 2, e1601039 (2016).

25. 25.

Dalby, M. J., Gadegaard, N. & Oreffo, R. O. Harnessing nanotopography and integrin–matrix interactions to influence stem cell fate. Nat. Mater. 13, 558–569 (2014).

26. 26.

Tian, B. et al. Coaxial silicon nanowires as solar cells and nanoelectronic power sources. Nature 449, 885–889 (2007).

27. 27.

Ghezzi, D. et al. A polymer optoelectronic interface restores light sensitivity in blind rat retinas. Nat. Photon. 7, 400–406 (2013).

28. 28.

Walter, M. G. et al. Solar water splitting cells. Chem. Rev. 110, 6446–6473 (2010).

29. 29.

Kang, D. et al. Electrochemical synthesis of photoelectrodes and catalysts for use in solar water splitting. Chem. Rev. 115, 12839–12887 (2015).

30. 30.

Merrill, D. R., Bikson, M. & Jefferys, J. G. Electrical stimulation of excitable tissue: design of efficacious and safe protocols. J. Neurosci. Methods 141, 171–198 (2005).

31. 31.

Ziegenfuss, J. S. et al. Draper-dependent glial phagocytic activity is mediated by Src and Syk family kinase signalling. Nature 453, 935–939 (2008).

32. 32.

Yoon, J., Park, J., Choi, M., Choi, W. J. & Choi, C. Application of femtosecond-pulsed lasers for direct optical manipulation of biological functions. Ann. Phys. 525, 205–214 (2013).

33. 33.

White, J. A., Blackmore, P. F., Schoenbach, K. H. & Beebe, S. J. Stimulation of capacitative calcium entry in HL-60 cells by nanosecond pulsed electric fields. J. Biol. Chem. 279, 22964–22972 (2004).

34. 34.

Hua, W., Young, E. C., Fleming, M. L. & Gelles, J. Coupling of kinesin steps to ATP hydrolysis. Nature 388, 390–393 (1997).

35. 35.

Stout, C. E., Costantin, J. L., Naus, C. C. & Charles, A. C. Intercellular calcium signaling in astrocytes via ATP release through connexin hemichannels. J. Biol. Chem. 277, 10482–10488 (2002).

36. 36.

Guthrie, P. B., Knappenberger, J., Segal, M., Bennett, M. V., Charles, A. C. & Kater, S. B. ATP released from astrocytes mediates glial calcium waves. J. Neurosci. 19, 520–528 (1999).

37. 37.

Chen, H. X. & Diebold, G. Chemical generation of acoustic waves: a giant photoacoustic effect. Science 270, 963–966 (1995).

38. 38.

Tang-Schomer, M. D., Patel, A. R., Baas, P. W. & Smith, D. H. Mechanical breaking of microtubules in axons during dynamic stretch injury underlies delayed elasticity, microtubule disassembly, and axon degeneration. FASEB J. 24, 1401–1410 (2010).

39. 39.

Maya-Vetencourt, J. F. et al. A fully organic retinal prosthesis restores vision in a rat model of degenerative blindness. Nat. Mater. 16, 681–689 (2017).

40. 40.

Ghezzi, D. et al. A hybrid bioorganic interface for neuronal photoactivation. Nat. Commun. 2, 166 (2011).

41. 41.

Lorach, H. et al. Photovoltaic restoration of sight with high visual acuity. Nat. Med. 21, 476–482 (2015).

42. 42.

Katz, L. C. & Dalva, M. B. Scanning laser photostimulation: a new approach for analyzing brain circuits. J. Neurosci. Methods 54, 205–218 (1994).

43. 43.

Yamawaki, N., Suter, B. A., Wickersham, I. R. & Shepherd, G. M. Combining optogenetics and electrophysiology to analyze projection neuron circuits. Cold Spring Harb. Protoc. https://doi.org/10.1101/pdb.prot090084 (2016).

44. 44.

Paralikar, K. J., Rao, C. R. & Clement, R. S. New approaches to eliminating common-noise artifacts in recordings from intracortical microelectrode arrays: inter-electrode correlation and virtual referencing. J. Neurosci. Methods 181, 27–35 (2009).

45. 45.

Veerabhadrappa, R. et al. Unified selective sorting approach to analyse multi-electrode extracellular data. Sci. Rep. 6, 28533 (2016).

46. 46.

Rossant, C. et al. Spike sorting for large, dense electrode arrays. Nat. Neurosci. 19, 634–641 (2016).

47. 47.

Yamawaki, N., Borges, K., Suter, B. A., Harris, K. D. & Shepherd, G. M. A genuine layer 4 in motor cortex with prototypical synaptic circuit connectivity. Elife 3, e05422 (2014).

48. 48.

Li, X., Yamawaki, N., Barrett, J. M., Kording, K. P. & Shepherd, G. M. G. Corticocortical signaling drives activity in a downstream area rapidly and scalably. Preprint at https://doi.org/10.1101/154914 (2017).

49. 49.

Tennant, K. A. et al. The organization of the forelimb representation of the C57BL/6 mouse motor cortex as defined by intracortical microstimulation and cytoarchitecture. Cereb. Cortex 21, 865–876 (2011).

50. 50.

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

51. 51.

Dai, J., Brooks, D. I. & Sheinberg, D. L. Optogenetic and electrical microstimulation systematically bias visuospatial choice in primates. Curr. Biol. 24, 63–69 (2014).

52. 52.

Stauffer, W. R. et al. Dopamine neuron-specific optogenetic stimulation in rhesus macaques. Cell 166, 1564–1571 (2016).

53. 53.

Chen, R., Canales, A. & Anikeeva, P. Neural recording and modulation technologies. Nat. Rev. Mater. 2, 16093 (2017).

54. 54.

Voyles, P. M., Grazul, J. L. & Muller, D. A. Imaging individual atoms inside crystals with ADF-STEM. Ultramicroscopy 96, 251–273 (2003).

55. 55.

Voyles, P. M., Muller, D. A., Grazul, J. L., Citrin, P. H. & Gossmann, H. J. Atomic-scale imaging of individual dopant atoms and clusters in highly n-type bulk Si. Nature 416, 826–829 (2002).

56. 56.

Suter, B. A. et al. Ephus: multipurpose data acquisition software for neuroscience experiments. Front. Neural Circuits 4, 100 (2010).

57. 57.

Yao, J., Liu, B. & Qin, F. Rapid temperature jump by infrared diode laser irradiation for patch-clamp studies. Biophys. J. 96, 3611–3619 (2009).

## Acknowledgements

This work is supported by the Air Force Office of Scientific Research (AFOSR FA9550-14-1-0175, FA9550-15-1-0285), the National Science Foundation (NSF CAREER, DMR-1254637; NSF MRSEC, DMR 1420709), the Searle Scholars Foundation and the National Institutes of Health (NIH NS101488, and NS061963). This work made use of the Japan Electron Optics Laboratory (JEOL) JEM-ARM200CF and JEOL JEM-3010 TEM in the Electron Microscopy Service of the Research Resources Center at the University of Illinois at Chicago (UIC). The acquisition of the UIC JEOL JEM-ARM200CF was supported by a MRI-R2 grant from the National Science Foundation (DMR-0959470). The animal imaging work conducted at the Integrated Small Animal Imaging Research Resource (iSAIRR) at the University of Chicago was supported in part by funding provided by the Virginia and D. K. Ludwig Fund for Cancer Research via the Imaging Research Institute in the Biological Sciences Division, by the University of Chicago Comprehensive Cancer Center including an NIH grant P30 CA14599, and by the Department of Radiology. Part of the schematic in Fig. 5c was generated from a three-dimensional anatomy software purchased from https://biosphera.org. The authors thank L. Yu, V. Sharapov, S. Patel, Y. Chen, Q. Guo and J. Jureller for providing technical support.

## Author information

### Author notes

1. These authors contributed equally: Yuanwen Jiang, Xiaojian Li, Bing Liu.

### Affiliations

1. #### Department of Chemistry, University of Chicago, Chicago, IL, USA

• Yuanwen Jiang
• , Jaeseok Yi
• , Kelliann Koehler
• , Vishnu Nair
• , Yin Fang
• , George Freyermuth
•  & Bozhi Tian
2. #### The James Franck Institute, University of Chicago, Chicago, IL, USA

• Yuanwen Jiang
• , Jaeseok Yi
• , Yin Fang
• , Xiang Gao
• , Kelliann Koehler
• , Vishnu Nair
•  & Bozhi Tian
3. #### Department of Physiology, Feinberg School of Medicine, Northwestern University, Chicago, IL, USA

• Xiaojian Li
• , KuangHua Guo
•  & Gordon M. G. Shepherd

• Bing Liu
5. #### The Research Resources Center, University of Illinois at Chicago, Chicago, IL, USA

• Fengyuan Shi
•  & Alan W. Nicholls
6. #### Department of Physics, University of Chicago, Chicago, IL, USA

• Edward Sudzilovsky
7. #### The Graduate Program in Biophysical Sciences, University of Chicago, Chicago, IL, USA

• Ramya Parameswaran
8. #### Ben May Department for Cancer Research, University of Chicago, Chicago, IL, USA

• Jiping Yue
•  & Xiaoyang Wu
9. #### Department of Radiology, University of Chicago, Chicago, IL, USA

• Hsiu-Ming Tsai
• , Chien-Min Kao
•  & Chin-Tu Chen
10. #### University Research Facility in Behavioral and Systems Neuroscience (UBSN), Hong Kong Polytechnic University, Kowloon, Hong Kong

• Raymond C. S. Wong

• Bozhi Tian

### Contributions

Y.J. and B.T. conceived the idea and designed the experiments. Y.J. fabricated the materials/devices with assistance from J. Yi, Y.F. (affiliation 2) and R.C.S.W.; X.L., B.L. and K.G. performed the brain slice and in vivo studies; X.L. and B.L. built the instrument and developed the software for in vivo neurophysiology experiments and analyses. Y.J., X.G., E.S., R.P., J. Yue, G.F. and X.W. performed the cell studies; Y.J., J. Yi, F.S., K.K., V.N., Y.F. (affiliation 1), H.-M.T., C.-M.K., C.-T.C. and A.W.N. performed the materials and biointerfaces characterizations; Y.J. developed the photoresponse analysis matrix and performed the COMSOL simulation; Y.J., X.L., B.L. and B.T. wrote the paper, and received comments and edits from all authors; B.T. and G.M.G.S. mentored the research.

### Competing interests

The authors declare no competing interests.

### Corresponding author

Correspondence to Bozhi Tian.

## Supplementary information

1. ### Supplementary Information

Supplementary figures, tables, video captions and references.

3. ### Supplementary Video 1

Left forelimb movement triggered by the photostimulation of a Si mesh (~4 mW for 50 ms).

4. ### Supplementary Video 2

Left forelimb movement triggered by the photostimulation of a Si mesh (~5 mW for 50 ms).

5. ### Supplementary Video 3

Right forelimb movement triggered by the photostimulation of a Si mesh (~5 mW for 50 ms).

6. ### Supplementary Video 4

Right forelimb movement triggered by the photostimulation of a Si mesh (~5 mW for 100 ms).

### DOI

https://doi.org/10.1038/s41551-018-0230-1