2.1.

Fabrication

2.1.1.

Neural probes with electrodes: design and fabrication

Neural probes with a high density of electrodes are shown in Fig. 1. The effective shank is $900\text{-}\mu \mathrm{m}$ long, $47\text{-}\mu \mathrm{m}$ wide, and $16\text{-}\mu \mathrm{m}$ thick [Fig. 1(a)]. Each shank hosts 64 electrodes with interconnections on a single layer. To accommodate such a high density of connections without using multilayer processes, electron-beam lithography, metal evaporation, and lift-off techniques are utilized. Wires are as small as 120 nm in width and 450 nm in pitch [Fig. 1(b)]. Wires can also be further scaled down to 65 nm in width and 240 nm in pitch [128 electrodes can fit inside a $45\text{-}\mu \mathrm{m}$-wide shank with a single interconnection layer, see Fig. 1(c)]. Electrodes are $5\text{-}\mu \mathrm{m}$ wide and $25\text{-}\mu \mathrm{m}$ long, with a spacing of $2.5\mu \mathrm{m}$. The fabrication process (Fig. 2) uses silicon substrates with a double-sided polished Si substrate with 100-nm ${\mathrm{Si}}_3{\mathrm{N}}_4$ layer, as a KOH etch mask [Fig. 2(a)]. On the wafer’s frontside, a positive photoresist, ZEP 520 A, is spun at 4000 rpm. and exposed at 100 kV with a Vistec VB300 electron-beam lithography system. After exposure, the resist is developed in amyl acetate for 1 min, using ultrasonication to improve resist contrast [Fig. 2(b)]. Electrical circuits are, therefore, defined by evaporation of 10 nm of titanium and 100 nm of gold [Fig. 2(c)] and then by a lift-off process (PG remover 1 h, 80°C) [Fig. 2(d)]. No sonication is utilized during the lift-off process to reduce the interconnection defectivity. A thin ${\mathrm{SiO}}_2$ layer (75 nm) passivates the interconnecting wires [Fig. 2(e)]. We deposit such a layer with an atomic layer deposition (ALD) tool (Oxford FlexAl, Plasma-Enhanced ALD, at 40°C). The ALD allows obtaining very high conformal, low stress, and thin passivation layers. To open the passivation layer on the electrodes, another electron-beam exposure (same parameters as the previous one) and a dry etching process are performed. This last step is done using a reactive ion etcher (RIE) tool (Oxford Plasmalab 80, using 35-sccm ${\mathrm{CHF}}_3$ and 25-sccm Ar, with $\mathrm{RF}=100\mathrm{W}$ and 30 mTorr of pressure). Thin membranes (15- to $25\text{-}\mu \mathrm{m}$ thick) are defined on the wafer’s backside [Fig. 2(f)] using the following processes. First, a UV photolithographic step (positive photoresist, MA-P 1215) defines the mask for the backside silicon nitride etching. Then, the backside nitride is etched (RIE, ${\mathrm{CHF}}_3$, and ${\mathrm{O}}_2$ with $\mathrm{RF}=100\mathrm{W}$ and 55 mTorr of pressure). The wafer front is afterward protected with ProTek B3.²⁰ ProTek B3 allows protecting the metal lines on the wafer’s frontside during the subsequent backside silicon etching. The thin membranes are finally obtained by wet etching the silicon (33% KOH at 80°C, for 6 h) on the wafer’s backside wherever the nitride mask had been previously removed. Once thin membranes are obtained, the ProTek layer on the front side is removed (2 h in Protek Remover 110).²⁶ Probes are defined onto the thin membranes by dry etching techniques onto the wafer’s frontside [Fig. 2(g)]. A 180-nm chromium mask is deposited by electron-beam evaporation, and then a photolithographic step defines the probe shape with a $100\text{-}\mu \mathrm{m}$-wide trench. The pattern is transferred to the chromium (wet etching in chromium etchant 1020, 4 min) and ${\mathrm{Si}}_3{\mathrm{N}}_4$ layers (dry etching with RIE) and finally in the silicon with an inductively coupled plasma (ICP) RIE (Oxford PlasmaLab 150 ICP, using 44-sccm ${\mathrm{SF}}_6$ and 6 sccm of ${\mathrm{O}}_2$ at $-120°\mathrm{C}$) to make free-standing probes. This last step is performed at cryogenic temperature to achieve straight sidewalls.²⁷ Free-standing probes are then soaked in chromium etchant 1020 to remove the residual chromium layer.

Fig. 1

Neural probes with high-density array of electrodes. (a) SEM image showing a zoom on device shank (thickness: $16\mu \mathrm{m}$ and width: $47\mu \mathrm{m}$) and the 64 electrodes. Scale bar: $20\mu \mathrm{m}$. (b) Zoom on interconnections (120-nm wide and with 450-nm pitch), passivated with 75 nm of ${\mathrm{SiO}}_2$ (darker edge around interconnections). Scale bar: 200 nm. (c) View of wires (65-nm wide and 240-nm pitch). A total of 128 electrodes are defined on a single layer. Scale bar: $10\mu \mathrm{m}$.

Fig. 2

Fabrication process of neural probes with high density of electrodes. (a) Starting wafer, (b) electron-beam exposure to define electrodes and interconnections, (c) titanium and gold evaporation, (d) lift-off in hot solvent, (e) deposition and etch after electron-beam lithography to open the passivation layer (75-nm ${\mathrm{SiO}}_2$, ALD), and (f) backside etch to thin the probes. The silicon is etched in 33% KOH. (g) The probe shape is then defined by front side dry etch.

2.1.2.

Neural probes with optical components: design and fabrication

${\mathrm{Si}}_3{\mathrm{N}}_4-{\mathrm{SiO}}_2$ waveguide elements can propagate waves in the visible wavelength range, where opsin activation takes place, with low absorption losses. Neural optical probes are designed with a main input single-mode waveguide, with core cross section of $250\mathrm{nm}\times 160\mathrm{nm}$, and two focusing gratings (one at half-length of the shank and the other near the tip). Focusing gratings are diffractive optical elements that can extract the light from a waveguide and focus it perpendicularly above the plane. Neural probes with optical elements are fabricated using a similar substrate and fabrication process as the probes with electrodes. Lionix TriPleX wafers²⁸ are made of a stack of silicon ($525\mu \mathrm{m}$), double-side ${\mathrm{SiO}}_2$ [$2.5\mu \mathrm{m}$, low pressure chemical vapor deposition (LPCVD) by Lionix] and ${\mathrm{Si}}_3{\mathrm{N}}_4$ (160 nm, LPCVD, by Lionix). Such wafers have been chosen since they present low absorption losses. An electron-beam lithography step (ZEP 520 A) defines the mask for waveguides and diffraction gratings [Fig. 3(a)]. The development is performed at room temperature in amyl acetate without sonication. The pattern is then transferred to the silicon nitride [Fig. 3(b)] (RIE, 48 sccm of ${\mathrm{CHF}}_3$ and 2 sccm of ${\mathrm{O}}_2$, with a pressure of 55 mTorr and an RF of 50 W) and covered with a $1.9\text{-}\mu \mathrm{m}$ ${\mathrm{SiO}}_2$ cladding layer [plasma enhanced chemical vapor deposition (PECVD) at 150°C, using 850-sccm 1% ${\mathrm{SiH}}_4/\mathrm{Ar}$ and 710-sccm ${\mathrm{N}}_2\mathrm{O}$]. Probes are thinned by backside etching [Fig. 3(c)] to form thin membranes ($55\text{-}\mu \mathrm{m}$ thick). First, the nitride and oxide backside layers are removed through dry etching. A chromium layer (180 nm) is evaporated on the wafer’s backside and is used as a mask for the dry etching of the nitride and oxide layers on the wafer’s backside. The chromium mask is defined by optical photolithography (MA-P 1215) and wet chromium etching. The silicon nitride is etched in an RIE tool (same as the electrical version), and the ${\mathrm{SiO}}_2$ layer is etched at $-100°\mathrm{C}$ in an Oxford Plasma lab 100 Viper (using a mix of 180-sccm ${\mathrm{SF}}_6$ and 20-sccm Ar, with 400 very high frequency and 250 RF). The silicon is then thinned in a 33% KOH bath (at 85°C) (the wafer’s frontside is protected, during the silicon etching, by ProTEK) and the ProTEK is finally removed. The probe shapes are defined from the wafer’s frontside [Fig. 3(d)] by etching the ${\mathrm{SiO}}_2$ (ICP, ${\mathrm{SF}}_6$ and ${\mathrm{O}}_2$ at $-100°\mathrm{C}$), then the ${\mathrm{Si}}_3{\mathrm{N}}_4$ (RIE, using ${\mathrm{CHF}}_3$ and ${\mathrm{O}}_2$), again the ${\mathrm{SiO}}_2$ (ICP, ${\mathrm{SF}}_6$ and ${\mathrm{O}}_2$ at $-100°\mathrm{C}$), and the remaining silicon (ICP, ${\mathrm{SF}}_6$ and ${\mathrm{O}}_2$ at $-120°\mathrm{C}$). The process uses a chromium mask (250 nm) patterned with optical photolithography and etched in chromium etchant 1020. The remaining chromium is removed by soaking the obtained free-standing probes in chromium wet etchant.

Fig. 3

Fabrication process of neural probes with optical elements. (a) Electron-beam exposure to define waveguides and gratings, (b) ${\mathrm{Si}}_3{\mathrm{N}}_4$ etching and cladding deposition (${\mathrm{SiO}}_2$, PECVD), (c) wafer’s backside etch in 33% KOH, and (d) dry etch to define probes on the front side.

2.1.4.

Electro-optical neural device integration

As stated above, the final goal is to be able to read out and manipulate single neurons activity in complex networks using simultaneously electrical and optical sensors. With respect to this, it is worth noting that (1) both electrical probes and optical probes use similar substrates (waveguides and electrodes are fabricated on silicon nitride) and (2) most of the fabrication steps are the same for both versions. An electro-optical probe will, therefore, have both layers, with waveguides buried below the electrodes [Figs. 4(a)–4(c)]. In particular, we want to have waveguides and electrodes separated by a $2.5\text{-}\mu \mathrm{m}$-thick ${\mathrm{SiO}}_2$ layer to keep light from scattering due to the gold interconnections. Electrodes and wires can cover all the optical features except for the focusing gratings. This method may allow achieving a simultaneous high density of electrode sites in combination with arbitrarily designed optical circuits. Optical circuits can be designed such that brain illumination is localized in the areas of interest. Light localization is obtained due to the gratings, which can be realized in multiple points along the shank. As an example, in Fig. 4(d), we show a design of optoelectronic probe with four buried optical waveguides each one having multiple gratings. A cross-section schematic is shown in Fig. 4(e)

Fig. 4

Schematic of the integrated optoelectronic probe. (a) Probe top view: the larger part has electrodes for wire bonding to the PCB and the waveguide input. The thinner part is the shank, which is inserted in the mice brain. (b) Zoom on the shank, showing electronic circuits (red: electrodes and blue: interconnections) and photonic circuits (in green). (c) Zoom near the tip, showing the grating. (a), (b), and (c) designs correspond, separately, to the fabricated electrical and optical probes. (d) Design example of a more complex photonic circuit with a high density of light output spots. Four waveguides are shown (although more could fit in the shank). Waveguides could be connected, for example, by splitters. (e) Cross section of (d). Electronic and photonic circuits are independent and can, therefore, occupy the entire shank area.

3.1.

Neural Probe Electrodes: Impedance Measurements and In Vivo Test

The neural probe is glued on the PCB and wire bonded [Fig. 5(a)]. Impedance is measured in saline with a NanoZ commercial system (Neuralynx) and shows an average of $4.5\pm 0.1\mathrm{M}\mathrm{\Omega }$ at 1000 Hz before electroplating. Electroplating in a black platinum solution (Neuralynx) is also achieved using the NanoZ by applying a DC current of $-0.1\mu \mathrm{A}$ for 12 s [Fig. 5(b)]. Electroplating lowers the average electrode impedance down to $200\pm 30\mathrm{K}\mathrm{\Omega }$ at 1000 Hz. In vivo experiments were conducted using head-fixed mice in the Adesnik Lab, UC Berkeley. Mice used for experiments were either wild type (ICR white strain from Charles River Laboratories) or Ai32-PV-Cre mice that express the excitatory opsin channelrhodopsin (ChR2). All experiments were performed in accordance with the guidelines and regulations of the Animal Care and Use Committee of the University of California, Berkeley (Protocol # AUP-2014-10-6832). A small craniotomy was made right above whisker somatosensory cortex (aka barrel cortex). This is where inputs from the whiskers cause neurons to spike vigorously to sensory contacts. The small probe shank was rigid enough to pierce the dura matter. While the mouse was running, a bar was placed in the sector swept out by the whiskers, which typically generates robust sensory activity in barrel cortex. After probe insertion and recording, spike sorting was conducted using the automated spike sorter Klusta. This spike sorter used probe geometry and an advanced algorithm to detect single units. After the automated step, we use the provided graphical user interface to manually curate the data. Spikes that have reasonably shaped waveforms, low refractory period violations, good separation in PCA space, and good autocorrelograms are accepted as coming from a single unit.²⁹ Main single-neuron waveforms for three adjacent electrodes are shown in Fig. 5(c).

Fig. 5

(a) The probe is glued on the PCB and wire bonded. Scale bar: 2 mm. (b) SEM image, showing electroplated pads on the shank. Scale bar: $10\mu \mathrm{m}$. (c) Plot of the mean (black) and standard deviation (gray) single-neuron waveforms recorded from three adjacent electrodes on the neural probe. For each of the three contacts, we plotted the largest amplitude of single-neuron signal recorded.

3.2.

Optical Neural Probe Light Focusing, Fiber Alignment, and Output Power

Neural probes with an input waveguide and two focusing gratings on the shank are fabricated, assembled, and characterized. One focusing grating is placed at half shank length and the other at $60\mu \mathrm{m}$ from the tip. Figure 6(a) shows a zoom on the last part of the shank, where one of the two output spots is (laser input in the waveguide: 520 nm). The waveguide input has a core cross section of 250 nm (width) and 160 nm (height). This ensures that only the fundamental mode propagates (waveguides are, therefore, single mode). At $500\mu \mathrm{m}$ from the input, the waveguide is tapered to a width of $4\mu \mathrm{m}$. Large waveguide width of $4\mu \mathrm{m}$ reduces sidewall scattering losses. The taper is $150\text{-}\mu \mathrm{m}$ long and allows for adiabatic mode conversion between narrow and wide waveguide sections. Single-mode regime is necessary to avoid multimode interference in the focusing gratings. Focusing gratings are diffractive optical elements that reconstruct the focus in a spot above the light propagation plane. These diffractive elements act like an integrated optical diffraction lens: light propagating in the waveguide is diffracted by chirped and curved grooves that deflect light rays at different angles; the light ray intersection point is the focal point.³⁰ The beam shape consists, therefore, of a light cone. The maximum of intensity of the beam is far from the shank plane (at the focusing height, which depends on the initial design of the grating), where neural cell density is reduced due to tissue damage during device insertion (the kill zone).³¹ Such illumination may activate, in the light cone, one or more neurons. The focusing height and beam angle depend on the initial design of the grating. Gratings are scripted in MATLAB according to the formula presented in Ref. 32. Different wavelengths can be extracted for a fixed grating design. As shown in Ref. 30, a different wavelength will result in a different diffraction angle and beam focusing height. Different wavelengths can be, therefore, employed to activate different optogenetic actuators.³³ As an example, we couple, in the same probe shown in Fig. 6(a), other wavelengths [Figs. 6(b) and 6(c): $520\mathrm{nm}+650\mathrm{nm}$; Figs. 6(d) and 6(e): $450\mathrm{nm}+650\mathrm{nm}$]. Different types of beam shape can be obtained by choosing a different design for the grating. For example, a grating with parallel lines would collimate the output beam (with small divergence angle). Different types of gratings can be designed based on the in vivo experiments results. Future in vivo tests will allow engineering the optimal spot size, focal height, and optimal input power to trigger action potentials.

Fig. 6

(a) Neural probe with a waveguide and two focusing gratings. The image shows the grating near the tip. Light input: 520 nm. Scale bar: $15\mu \mathrm{m}$. (b) Zoom on the tip. Laser input: $520\mathrm{nm}+650\mathrm{nm}$. Scale bar: $30\mu \mathrm{m}$. (c) Zoom on the focusing grating output spots. Scale bar: $4\mu \mathrm{m}$. (d) Probe tip with laser input $450\mathrm{nm}+650\mathrm{nm}$. Scale bar: $30\mu \mathrm{m}$. (e) Zoom on the focusing grating output spots. Scale bar: $4\mu \mathrm{m}$.

Waveguide losses are measured using test structures to be $8.1\mathrm{dB}/\mathrm{cm}$ for blue light (450 nm) and $5.4\mathrm{dB}/\mathrm{cm}$ for red light (650 nm). We designed two types of focusing gratings, one for red light (650 nm, focusing height: $10\mu \mathrm{m}$) and one for blue light (450 nm, focusing height: $20\mu \mathrm{m}$). Focusing grating output spots are imaged by a CCD camera and measured at different heights with respect to the grating plane. Measurements are performed in air. For the 650-nm grating, at the height of $10\mu \mathrm{m}$ (the grating focusing distance), the spot intensity is maximum and its full width at half maximum (FWHM) is measured to be $1.65\mu \mathrm{m}$ (data not shown). For the 450-nm grating, at the focusing height of $20\mu \mathrm{m}$, the FWHM is $2.35\mu \mathrm{m}$ [Fig. 7(a)]. The grating lens is shown in Fig. 7(b). Images of spots at different CCD heights can be seen in Fig. 7(c). The majority of losses, estimated to be around 33 dB, are at the probe base (so far from the shank), where the fiber is aligned to the input waveguide on the probe. These high losses are due to a nonoptimal fiber to waveguide coupling (fiber misalignment during glue curing). Focusing grating output coupling efficiency is estimated to be between 3.5 and 6 dB. The efficiency varies with the wavelength due to interferences between light extracted upward by the grating and light extracted downward and reflected at the bottom oxide cladding–silicon interface. To activate channelrhodopsins, a threshold of $1.5\mathrm{mW}/{\mathrm{mm}}^2$³⁴ is required. This translates to a power density of $1.5\mathrm{nW}/{\mu \mathrm{m}}^2$. The spot intensity on the neural probe is calculated after aligning the fiber and gluing it. Once glued, the fiber does not move, and therefore, the probe output power does not change for a fixed laser input power. The intensity value in the focusing point, with a laser input of 2 mW, equals 35.5 nW at half shank and 8.75 nW for the grating near the tip. This corresponds, at the focusing height (spot area: $5.5{\mathrm{mm}}^2$), to a power density of $6.43\mathrm{mW}/{\mathrm{mm}}^2$ for the grating at half shank and $1.58\mathrm{mW}/{\mathrm{mm}}^2$ for the grating near the tip. These results highlight the possibility of achieving neuron light activation that occurs within close proximity to the focused spot. The volume pertaining to the light cone, where channelrhodopsins can be activated, can be enlarged by increasing the laser power.

Fig. 7

(a) Relative intensity of the output spot of the probe focusing grating. The grating shown is been designed for blue light (450 nm), with a focusing height of $20\mu \mathrm{m}$ above the device plane. The intensity is plotted as function of a line along the focusing grating direction. The three curves show the image intensity recorded at different heights with respect to the grating’s plane ($h=0$ means the camera focus lies on the grating and $h=10$ means the camera focus is $10\mu \mathrm{m}$ above the grating plane); error bars for the vertical axis correspond to noise (in plane scattered light). Error bars on the horizontal axis correspond to the pixel size. (b) SEM image of the probe focusing grating. (c) Output spot images at different camera heights. Left: CCD camera focused on grating plane. Center: CCD camera focused $20\mu \mathrm{m}$ above the grating plane (the focal plane). Right: CCD camera focused $100\mu \mathrm{m}$ above the grating plane; after the focusing point, the beam broadens.