First, we defined the following as variable parameters of the biphasic current pulse stimulus: the amplitude and duration of each of the cathodic and anodic phases, the temporal order of the phases (i.e., cathodic-first or anodic-first), the inter-phase interval, and the inter-pulse interval. Based on previous studies on the chronaxie (1, 29–31) and the charge threshold for inducing phosphenes (13, 14, 16, 29) or neural excitations (27, 28), the temporal accuracy and the amplitude resolution for controlling those variable parameters of stimulus were roughly determined to be less than several micro-seconds and a few micro-amperes, respectively. Also, we assumed that the pulse repetition frequency of stimulation (i.e., the inverse of the inter-pulse interval) was less than 300 Hz (30, 41). Therefore, if a single pulse of the biphasic current stimulus has the duration of around 0.4 msec, for instance (30), then eight stimulus pulses can be accommodated within such an inter-pulse interval. In addition, the spatial pattern of stimuli should better be updated at 30-50 frames per second (42, 43). These numerical conditions were considered for designing the stimulation module, as described below.

Figure 1A shows the main block diagram of our stimulation module as an ASIC chip. Given the chip die size of about 4-by-4 mm, we arranged eight current generators (“G#1” to “G#8” in the figure) together with other circuit blocks in the chip. Based on the above-mentioned consideration of the inter-pulse interval, each of those current generators has eight channels for the stimulus output. Thus, stimulus current pulses can be generated from 8 channels in parallel, or 64 channels at maximum in time-division semi-parallel mode. For use of the time division multiplexing with the 1:8 configuration (i.e., 8 output channels per current generator), the maximum rate of the stimulus pulse repetition in a given output channel is limited by the total length of time required for completing all stimulus outputs from the 8 output channels in a given current generator. The shortest time required for multiplex switching from one to the next output channel is ~7.8 μsec in the present design. Thus, if the time length of every biphasic stimulus pulse is 0.4 msec for instance, then the maximum pulse repetition rate in a given output channel to be ~306 Hz. For controlling the stimulus outputs from the 64 channels, on-chip register memories are implemented (44). Namely, each of the current generators contains a set of data registers (“R#1” to “R#8” in the figure) that stores the stimulation parameters for every output channel, including the current amplitude, the sequential order of output from the channels, and the stimulus on/off switching. In addition, a pair of data registers (“R#L” and “R#R” in the figure) sets the circuit operation parameters, such as maximum current output range, use of the inrush current suppression (described later), and bias voltages in the current generator circuits. Detailed information on these registers, such as hardware locations, register names, data contents, and data sizes, is listed in Table 1. To reduce the overhead time for storing the data into these registers, they are divided into three groups depending on the frequency of data update (see the leftmost column in Table 1). Register group #1 (“Reg1,” 64 bits) manages the stimulus on/off switching and is updated synchronously with the spatial pattern of stimuli. Register group #2 (“Reg2,” 192 bits) is used to set the stimulation sequence and is updated at a moderate frequency. Register group #3 (“Reg3,” 1160 bits) consists of registers for the other stimulus and circuit parameters and is updated infrequently.

In addition to the register setting (“RS”) signals mentioned above, the digital control signals are also provided from an external device to the module chip via bus input ports (“RS” and “Control signals” in Figures 1A,B). Besides, an individual module chip is pre-assigned with an identification (ID) number using a 6-bit digital code set via another input port (“Chip ID” in Figure 1A). As illustrated in Figure 1B, multiple module chips (up to 2⁶ = 64) can be connected to the two common buses for the RS and Control signals. The 6-bit ID code is used as the header of the RS signal, and a chip selector circuit in each of the module chip enables updating of the registers when the header is matched with the pre-assigned ID number. After completing the register updates for all of the module chips, the stimulus outputs from all the module chips (up to 64 chips) are driven by the Control signals on the common bus. In other words, the control of a single chip or multiple chips is divided into two phases: initialization and stimulation. In the initialization phase, the data sets in the Reg2 and Reg3 are updated on a chip-by-chip basis. The time required for this phase is 136.2 μsec per chip and thus 8.717 msec for 64 chips at 10 Mbps (17, 45, 46). In the stimulation phase, the data sets in the Reg1 are also updated on a chip-by-chip basis, and then all of the module chips are operated to output the stimulus current pulses by the Control signals in a time-division semi-parallel mode with eight timeslots. The time durations required for updating the Reg1 are 7.3 μsec per chip and 467.2 μsec for 64 chips at 10 Mbps for instance (17, 45, 46). This overhead time is thought to be short enough to control the 64 chips and thus 4,096 stimulating electrodes while updating the stimulation position in the data in the Reg1 (i.e., the spatial patterns of stimuli) every 20 msec (i.e., 50 frame-per-second) for instance (42, 43).

Figure 1C shows the circuit diagram of the current generator, which is composed of an R-2R ladder-type current-mode D/A converter, a differential amplifier, pMOS and nMOS current mirrors, and output circuits (47). In addition, in each of the current mirrors, the gain boosting circuitry with a pair of auxiliary amplifiers is incorporated to realize high output impedance in the voltage range below ±2.42 V. The stimulus current amplitude can be set with a 6-bit code and the maximum output range of the current mirror output is selectable from approximately ±100 to ±400 μA by a ~±100-μA step (“max amplitude setting” in the figure). For instance, considering the threshold stimulus charge for inducing phosphenes or neural excitations (21, 30), the maximum output range of ~±100 μA would be selected as a default. In this range, the amplitude resolution is approximately 1.56 μA (i.e., 100 μA divided by 2⁶). In an actual fabricated chip of the module, however, the amplitude resolution varies between anodic and cathodic currents in a given current generator, and varies also among different current generators (but not among 8 output channels of the same current generator), mostly because of the difference in property of the current mirrors.

The output current flows into or out of either one of eight output circuit blocks (boxes in the rightmost of Figure 1C), each of which is connected to a stimulating electrode at its output port (from “ch#1” to “ch#8”). Thus, by controlling these output circuits, current stimulation is realized by time division with the eight stimulating electrodes. In general, when the surface potential of the electrode in contact with the tissue differs from the potential of the current mirror output, and if the output circuit is a simple analog switch, then an inrush current accompanying the potential difference occurs at the moment when the switch is connected. Therefore, as illustrated in the boxes in Figure 1C, a feedback buffer circuit is added to connect the circuit output and the current mirror output, and the potentials of the current mirror output and the electrode can be rendered equal until the instant before stimulation to suppress the inrush current.

We implemented also a circuit that enables fine adjustment to the pulse width of the second phase in a biphasic stimulus pulse (“pulse-width adjustment” in Figure 1C) for reducing the charge imbalance due to the nMOS/pMOS mismatch. The second phase duration can be either increased or decreased at ~1.95 μsec/step with a 4-bit code in every current generator. In addition, as shown in the boxes in Figure 1C, a switch for shorting of each output channel to a reference electrode/potential (“Vref” in the figure) in the tissue is also built into the output circuit such that charges accumulated at the electrode surface and in the tissue, as well as the residual DC current (48), can be further reduced through such a shorted connection. In practice, the amplitudes and the durations of both anodic and cathodic phases are first set, and then the pulse width adjustment would be used to minimize the charge imbalance for every current generator, while measuring the output current in a dry-bench test and/or a wet-bench test prior to an animal experiment. For certain animal experiments, this procedure can be enough for determining the stimulus charge with the best possible accuracy. However, the balancing of charge only with this procedure may not be perfect, and gradual charge accumulation might occur due to unpredictable factors in biological tissues and the electrode-tissue interface during a long-term stimulation in animals in vivo. In such a case, the shorting to Vref can reduce the charge accumulation to compensate for the imperfectness of the above-mentioned procedure. At least in some animal experiments, these two approaches, i.e., fine pulse-width adjustment and output shorting, suffice for suppression of the voltage accumulation during repetitive pulsing of the stimulus current (12), and may offer an alternative to attaining a perfect charge balance of every single stimulus pulse (49–54). Besides, these circuit operations are optional, allowing the user to intentionally test the possible effects of charge imbalance on neural responses or electrode/tissue damages in experimental animals.

Figure 1D shows the layout design of the module chip with a die size of 3.64 mm by 3.64 mm, in which the control circuit at the center and the eight current generators at the right and left sides are recognizable. This chip was fabricated via the 0.25-μm CMOS process (Taiwan Semiconductor Manufacturing Co., Ltd), packaged and mounted on a printed circuit board (PCB) (Figure 1E).

Basic operations of the module were bench tested. In those tests, the 6-bit high/low voltage code for the “Chip ID” was supplied from the power line on the PCB. The RS and Control signals fed to the chip were generated by either one of a portable pattern generator (UPG-2116, Japan Data Systems Inc., Hyogo, Japan), an FPGA integration module (Spartan-6, Xilinx, CA, U.S.A.; XEM6010-LX150, Opal Kelly Inc., OR, USA), or a prototype wireless module system (55). Some examples of the results of dry-bench tests are described below.

First, the power consumption of the module chip was measured. The SPICE simulation with the present chip design produced an estimated power consumption of ~2.54 mW without stimulus output, and an actual power consumption measured in the dry-bench test was ~2.27 mW. When the number of the output channel in use was increased under the condition that the biphasic current pulses (~40 μA/phase in amplitude, 0.1 msec/phase in duration, 200 Hz in frequency) were passed for every output channel to a ~10-kΩ resistor, the actual power consumption increased by ~21 μW per channel. These actual power consumptions changed little (<0.5 %) when the clock frequency for generating the digital RS and Control signals was varied in the range of 1–12 MHz.

Second, the output DC impedance of the current generator in the module was examined. An Ampere meter (ammeter) and a voltage source (34405A and E3631A, Agilent Technologies Inc., CA, USA.) were connected in series to one of the output channels of the module. The voltage at the channel output was varied from zero to either +2.5 or −2.5 V by the voltage source while the command current amplitude of the current generator in the stimulation module was set to a constant value of either one of approximately ±10, ±20, ± 40, and ±80 μA, and the actual output current was measured by the ammeter. Figures 2A,B plots the measured output current against the voltage at the channel output. In either the anodic (A) or the cathodic (B) direction, the output current level remains constant when the output voltage is less than approximately ±2.4 volts, which is the limit voltage determined by the gain boosting circuitry in the current generator (see Figure 1C). The output current level varies less than 0.1 μA (the resolution of the ammeter used) in total when the output voltage level varies from 0 to ~±2.4 volts, indicating that the output DC impedance is higher than 24 MΩ. This was the case in all the 8 current generators, and hence all the 64 output channels.

Third, the pulse output from the current generator in the module was examined with using practical impedance loads. The parallel circuit of a resistor and a capacitor was used as a model of the electrode-electrolyte interface (11), and was connected in series with a ~1-kΩ resistor (“Rs”) to one of the output channels of the module (Figure 2C). Two model circuits were used in this test example; one mimicked a metal-based electrode usable for intracortical stimulation (16) (~33.7 kΩ at 1 kHz), and another mimicked a similar electrode but with much smaller surface area (~325 kΩ at 1 kHz). Figure 2D shows the current traces recorded as voltage drops across the resistor Rs (“Iout” in Figure 2C) and Figure 2E shows the resulting voltages recorded with respect to the voltage reference (“Vout” in Figure 2C). The waveforms of the current pulses are nearly the same (red and green traces in D) regardless of the difference in voltage drop (red and green traces in E) between the two model circuits. As expected, when the output voltage exceeds the limit voltage of ~±2.4 V, the current output from the generator is limited (Figures 2F,G). Namely, at the amplitude setting of ~±80 μA, the current amplitude during the cathodic phase was first kept near the command value, but subsequently decreased to a smaller value as the output voltage reached the limit voltage. The similar tests were conducted with using other output load impedance ranging from ~8 to ~513 kΩ at 1 kHz. The 10–90% rise time was equal to or shorter than 5 μsec, and the current amplitude during the cathodic and anodic phases were fluctuated by less than 10 %, as far as the output voltage were below the limit value and as the output impedance was lower than around ~150 kΩ at 1 kHz (not shown). We also verified that the current outputs greater than ±100 μA/phase in amplitude were able to be generated when a 1-kΩ resistor was connected as the output load (Supplementary Figure 1).

Fourth, we verified that the current pulse outputs from all the 64 output channels of the module can be operated in a time-division semi-parallel manner. A ~10-kΩ resister was connected to every output channel and the current output was recorded as voltage drop across the resistor. Figure 3 shows the traces of those current outputs from all the 64 channels. In this recording, a time window of ~4,936 μsec was divided into 8 timeslots and, in each of those time slots, every current generator (from “G#1” to “G#8”) generated the biphasic current pulse with one out of the 8 output channels (from “ch#1” to “ch#8”). The register setting values for the anodic and cathodic amplitudes (i.e., AP and CP in Table 1) of every current generator were selected so that the actual amplitudes in all the 64 output channels were approximately ±30 μA. In the figure, each row overlays 8 traces of the current outputs from the 8 output channels of a particular current generator, and the 64 traces on the same timeline are shown. The onset and offset timings as well as the amplitudes of these current pulses little fluctuated when the same pattern of the outputs were repeated, showing the reliable operation of the time-division semi-parallel mode. Also, the time precision as well as the current amplitude accuracy were little affected by varying the number of the generators/output channels in use (not shown).

All animal care and experimental procedures in the present study were approved by the Animal Experiments Committee of Osaka University and conducted in conformity with the ‘Guidelines for Proper Conduct of Animal Experiments' by the Science Council of Japan. The Long-Evans rats (24 rats, 6 to 16 weeks old, male and female; Japan SLC Inc., Shizuoka, Japan) were used because of their enough area of the visual cortex for insertion of a multi-electrode array (MEA) (Figure 4G). During the surgery and the imaging experiment, the animal's rectal temperature was feedback-controlled at ~37 °C and electrocardiogram (ECG) was monitored. The surgical procedures and the dye staining were almost the same as those in the previous experiments in rats in vivo (26, 55). In brief, following administration of atropine (10 μg) and dexamethasone (20 μg), anesthesia was made with either (1) intraperitoneal (IP) injection of a mixture of medetomidine (0.15 mg/kg-b.w.), midazolam (2 mg/kg-b.w.) and butorphanol tartrate (2.5 mg/kg-b.w.), (2) IP injection of urethane (5 ml/kg-b.w. with 25 wt-% soln., or (3) isoflurane inhalation (0.5–4 % at ~0.5–2 l/min). Craniotomy was performed on the right posterior parietal bone (2–8 or 3–9 mm posterior and 1–7 mm lateral from the Bregma; blue and light blue dotted boxes in Figure 4F) and a custom-made chamber was attached on the bone around the craniotomy. The exposed cortex with the dura intact or removed partially was stained with VSD (RH1691 or RH2080; Optical Imaging, Ltd, Rehovot, Israel). The dye staining of layer II/III (57) was confirmed in a coronal section of the cerebrum after the imaging experiment.

Figure 4A illustrates the imaging setup (26, 55) combined with the stimulation module. Similar to the dry-bench test, the RS and Control signals fed to the stimulation module were generated by either the portable pattern generator or the prototype wireless system (see Section Dry-bench test). A custom-designed MEA (MicroProbes, Inc., MD, USA; Figures 4B–D), in which three to five needle-shaped iridium electrodes were aligned in line, was connected to the output channels of the stimulation module. In the example shown in the figure, the inter-tip distances were ~0.29–0.36 mm (C) and the electrically-exposed tips (asterisks in C and area with dotted outline in D) had conical shapes with base diameters of ~20 μm and lengths of roughly 100 μm (D). Before the experiment, the exposed tip of each electrode was activated so that the impedance was reduced to ~10–65 kΩ at 1 kHz as measured in an artificial cerebrospinal fluid (Figure 4E). The operations of the stimulation module and the imaging system were controlled with using a common trigger signal generated with reference to the ECG R-wave (arrows of “trigger” in Figure 4A). In some experiments, a commercially available desktop stimulator (STG2008 with the ±1.6-mA current range, MultiChannel Systems, GmbH, Germany) was also used for comparison. The imaging method was the same as that in the previous experiments (26, 55). In brief, the VSD fluorescence emitted from a 6.25 × 6.25-mm area on the cortex under the excitation illumination was captured at 1000 frames/s with a 100 × 100-pixel CMOS sensor (MiCAM-Ultima, BrainVision, Inc., Tokyo, Japan). The captured image stream was processed as described previously (26, 27) so that the pixel values of an image represent the fractional fluorescence changes induced by the stimulus (referred to as ΔF/F in the following text). In the ΔF/F image stream, the increase and decrease of pixel value correspond to relative depolarization and hyperpolarization, respectively, of the cytoplasmic membrane.

In order to identify the area of the primary visual cortex (V1), diffuse light stimulation (~13–16 photon/μm²/msec in intensity and 20–500 msec in duration; 525-nm LED as the source) was applied to the left eye via a light guide. At 20–40 msec after the stimulus onset, an increase in ΔF/F was initiated in a certain cortical area (color contour plot in Figure 4; also Figures 6B, 7B), the location of which agreed with the V1 location on the rat brain atlas (58, 59) (black outline in Figure 4F). Through the craniotomy window, the electrode tip was inserted in the V1 area to reach a depth of 300–600 μm from the surface (Figure 4G). The insertion angle was approximately 40° to the cortical surface plane. With this configuration, the cortical surface was only slightly obscured from sight by the electrode shafts of MEA under the epifluorescence microscope (see Figures 6A, 7A, 9A). A pellet or winding wire of Ag/AgCl was placed as the return electrode for stimulation in the head chamber, or in the mouth in a few experiments. The return electrode was connected the voltage reference port of the stimulation module or of the desk-top stimulator. The current pulse stimuli were delivered at inter-stimulus intervals of more than 10 sec to avoid any irreversible effects on the tissue and/or the response.

Although proper operations of the stimulation module as an electronic device were verified in the dry-bench test (Section Dry-bench test), yet the tests in animals in vivo were considered essential since the electrode-electrolyte interface as well as other current-passing pathway in the biological tissues could exhibit non-linear electro-chemical properties (60). Therefore, the current pulse outputs through the MEA inserted in the cortex in vivo was first examined. In addition to the stimulation module, the desktop stimulator was used as a reference in the same animals and MEAs. Figures 5A,B shows an example of the current pluses recorded as voltage drops across a 1-kΩ resistor between one of the stimulating electrodes and an output channel of the desktop stimulator (A) or the stimulation module (B). The command amplitude of the current pulse was varied in the range from approximately ±5 to ±30 μA by a ~5-μA step. With a given amplitude setting, the current traces were averaged across sixteen recordings to reduce noise. As shown in Figure 5A, the onset of the cathodic pulse with relatively small amplitude settings was delayed, and the rise and fall time at the onset and offset of the current pulses was 20 μsec or longer. These characteristics of the current outputs were within the specifications of the stimulator used here, thereby ensuring the practically adequate configuration of the electrode placement and the current-passing pathway in the animal. When the stimulator was replaced with the stimulation module (Figure 5B), the rise and fall time at the current pulse onset and offset was nearly equal to or shorter than 5 μsec, and the amplitude was kept stable in either the cathodic or anodic phase with every amplitude setting. This was similar also in other animals and electrodes (see Figure 5E and Supplementary Figures 2A,C), indicating the reliable control of stimuli by the module.

Figure 5C shows a plot of the charges delivered by the stimulus pulses in either the cathodic or the anodic phase. The measured values (i.e., the time-integral of the current trace) deviated slightly from the command values when the stimulation module was used (red dots in the plot). These deviations were similar between the negative and positive charges. Figure 5D shows the difference between the positive and negative charges (i.e., the imbalanced charge) plotted against the charge delivered by the cathodic pulse. Since the maximum output range of the module was set to be ~±100 μA and in turn the amplitude resolution was approximately 1.56 μA (see Section Design), the theoretical upper limit of the charge imbalance between the cathodic and anodic phases was ±156 pC. Consistently, the imbalanced charges (red dots in the plot) were smaller than this limit (indicated by hatching in the figure). Such a measurable amount of the charge imbalance could be reduced by utilizing the pulse-width adjustment in the stimulation module (see Section Design), as shown in Figures 5E,F. The current amplitude and the pulse duration were set to be ~±30 μA/phase and 0.2 msec/phase, respectively, as the command values. With these settings, the offset timing of the anodic phase was shifted from one to the next by using the pulse-width adjustment function, and the corresponding current traces were recorded, as shown in Figure 5E. In the figure, eleven traces are superimposed, and each of these traces represents a single recording sweep. The inset shows an expanded view of the current traces near the offset of the anodic phase. Figure 5F plots the imbalanced charge against the adjusted pulse width of the anodic phase. At the best setting in this example, namely a pulse width shorter than 200 μsec by ~1.95 μsec, the imbalanced charge was reduced marginally below +6 pC.

Finally, the usability of the stimulation module for the physiological experiments in vivo was examined. The same set of the current pulses shown in Figure 5 was used as the stimuli. Figure 6C shows an example of pseudo-color time-lapse images of the ΔF/F signal in response to the stimuli delivered via one of the electrodes in MEA. As compared in the top two rows of the figure, when the current amplitude was greater than ~5 μA/phase, the increase in ΔF/F corresponding to neural excitations appeared significant. The increase in ΔF/F was initially induced near the tip of the stimulating electrode (red line in the figure shows the outline of shadow of the electrode), and then propagated to its surrounding area in the next several milliseconds. The spatial extent of this propagation was larger as the current amplitude was increased from ~10 to ~20 μA/phase (the third and fourth rows from the top in the figure) and nearly saturated when the current amplitude was greater than ~20 μA/phase (two rows at the bottom in the figure). The saturation of the response was not due to a saturation of the output voltage in the stimulation module (Supplementary Figures 2A,B), and was similar when the desktop stimulator instead of the stimulation module was used in this particular animal and the electrode (Supplementary Figures 2C–F). Figure 6D shows the time courses of the changes in ΔF/F near the tip of the electrode. In response to the stimuli, the ΔF/F signal increased to reach the positive peak at 11–15 msec post-stimulus time, and then returned to the original level after a few to several hundred milliseconds. In the case that the current amplitude was greater than ~10 μA/phase, the ΔF/F signal decreased below the original level to reach the negative peak at around 100 msec post-stimulus time. These were consistent with those observed in the previous experiments (26, 27). Figure 6E plots the positive peak amplitudes of the ΔF/F responses against the stimulus charges. For this figure, the data sets obtained from multiple samples were averaged (n = 8 electrodes in 6 rats) and fitted with a modified Hill function (dotted blue line), providing approximate values of ~0.9 nC/phase for the threshold charge (I_th), ~2.4 nC/phase for the charge that induced the half-maximum response amplitude (I₅₀), and ~2.5 for the Hill coefficient (N).

Figure 7 shows an example of the multi-site neural excitations with using the stimulation module. Figure 7A shows the VSD fluorescence image of the exposed cortex superimposed with the locations of four electrodes of the MEA (colored outlines of shadows of the electrodes; Ch1, Ch2, Ch3, and Ch4). The current pulse with the amplitude of ~30 μA/phase was delivered from either one of those four electrodes (four rows from the top in Figures 7C,D). In each of the four cases, the increase in ΔF/F was initiated at around a few milliseconds post-stimulus time, as shown in Figure 7C. The center peak positions of these initial responses shifted from one another by around 0.3–0.4 mm within the V1, consistent with the tip positions of the four electrodes (outlines of shadows of the electrodes). In each of the four cases, the increase in ΔF/F propagated to its surrounding area in V1 in the next several milliseconds, and then further circuit activations were induced in V1 as well as extra-V1 areas in the next tens of millisecond, as shown in Figure 7D. The spatiotemporal patterns of those cortical activations differed from each other among the four cases. In the experiment shown in the bottom row in Figures 7C,D, the current pulses with the amplitude of ~10 μA/phase were delivered simultaneously from three of those electrodes (Ch1 to 3). As shown in the image frames at 3–4 and 5–6 msec post-stimulus time (Figure 7C), the spatial profile of the initial increase in ΔF/F was slightly more elongated in the anterior-posterior direction than those induced by the single-site stimulation. The spatiotemporal pattern of the cortical activation induced after this initial excitation appeared somehow similar to that induced by the stimulus with the electrode Ch2 (Figure 7D).

Figure 8 shows an example of the sequential multi-site neural excitations with the stimulation module. The current pulse with the amplitude of ~30 μA/phase was sequentially delivered from the electrode Ch1 to Ch 3. The time interval between the two pulses was set to be 20 msec (Figure 8A), since the previous physiological study showed that the inhibitory circuit is recruited in this time range after the preceding stimulus pulse (27, 61). As expected, the circuit activation seen after the second stimulus pulse decayed more rapidly (VSD images and spatial line profiles from 29 to 38 msec in Figures 8B,D, respectively) than that seen after the first stimulus pulse (VSD images and spatial line profiles from 9 to 18 msec in Figures 8B,D, respectively), and the neural excitation induced by the third stimulus pulse (bottom row in Figure 8C; green traces in Figures 8D,E) was significantly smaller in amplitude than those induced by the first and second stimulus pulses (top and middle rows in Figure 8C; purple and red traces in Figures 8D,E).

To our best knowledge, this is the first time that the spatio-temporal patterns of cortical neural responses to multi-channel stimulation at a millisecond time resolution was examined. The results demonstrated that the stimulation module can be used to induce the neural excitations in different spatial locations within V1, and to examine the non-linear interaction of those neural excitations and the subsequent circuit activations in the visual cortices. Further experiments are necessary to investigate quantitative relationships between the spatiotemporal patterns of the stimulation and the corresponding neural responses. Nevertheless, these results suggested that the stimulation module is useful for such physiological investigations.

We also examined the neural excitations induced by the bipolar stimulation (Figure 9). One of five MEA channels (Ch5) was electrically shorted to the Ag/AgCl reference electrode by the switch in the output circuit (see boxes in the rightmost of Figure 1C) to act as a return electrode for stimulation. Either one of the remaining four MEA channels (from Ch1 to Ch4) were paired with this return electrode to act as a working electrode for stimulation. The current pulse with the amplitude of ~30 μA/phase was delivered from either of those four pairs of electrodes. The same current pulse was also delivered either of the four working electrodes in the monopolar configuration. Figure 9B shows the VSD images in response to the bipolar stimulation, and Figure 9C compares the spatial line profiles of VSD signals in response to the bipolar stimulation (red traces) and the monopolar stimulation (blue dotted traces). When the inter-tip distance of the MEA pair was ~0.6 mm or longer (i.e., when either of Ch1, Ch2 and Ch3 was used as the working electrode), the neural excitations were initiated in two discernible locations (3–4 msec post-stimulus time in Figure 9B), and those were larger in amplitude and wider in spatial extent (red traces in Figure 9C) than those induced by the monopolar stimulation (blue traces in Figure 9C). When the inter-tip distance of the MEA pair was ~0.3 mm (i.e., when Ch4 was used as the working electrode), the spatial profile of the neural excitation was similar to that induced by the monopolar stimulation (the rightmost plots in Figure 9C).

This result demonstrated that the ability of the stimulation module to arrange the working-returning electrode combination by electronic means can be beneficial in modifying the spatial pattern of neural excitation while keeping the stimuli the same.