I N VITRO EXPERIMENT RESULTS

The main purpose of this in vitro experiment is to verify the electrical stimulation of the retinal chip and we provide the light-response experiment and electrical-response which the electrical simulation is from the retinal chip which is triggered by IR. In compare with the light-response and electrical-response, the light intensity of the CRT light stimulation (visible light) is too low to trigger the retinal chip and the waveform recorded is almost the pure light-response of the ganglion cell, and the IR stimulation won’t trigger the normal ganglion cell due to that IR is not visible light and recorded waveform is definitely not from the light-response of ganglion cell but from the electrical stimulation of the retinal chip which triggered by IR. The ganglion cell used in this in vitro experiment is shown in Fig. 3.6. Fig.

3.7 shows the extra-cellular signal which recorded under varying light intensity of IR stimulus.

Fig. 3.8 shows the zooming of the extra-cellular signals above. There is a spike with latency shown after the IR stimulation while the IR intensity is over 400mW. This spike is a strong evidence to prove the retina cell is triggered by the MPA. Fig. 3.9 shows the comparison between electrical stimulation and light-response. With the same experiment environment, including chip, retina, probe position and temperature, two responses from two different stimulations are recorded individually. Left block shows the signal which recorded while silicon retina was triggered by IR irradiation. Right block shows the ganglion cell response while the visible light triggers the retina. Despite the artifact, the latency response of both signals has some in common. This data shows that the sub-retinal prosthesis with silicon retina which powered by solar cell is able to trigger the retina and may replace the functionalities of outer-plexiform layer of the retina.

In order to compare with the Fig.3.6, Fig. 3.10 shows the extra-cellular signal which recorded under varying light intensity of IR stimulus but the retina is removed. As shown in Fig.3.11, with the same experiment environment, including chip, IR stimulation, probe position and temperature, two signals are recorded individually. Left block shows the signal which the retina is still attach on the silicon retina. Right block shows the signal while the retina is removed from the chip. Under the same IR stimulation, the right block show some artifacts which is induced by the silicon retina. The same artifact is also shown in left block, but the spike with latency is clearly shows the feasibility of the silicon retina which can generate effective stimulation under IR stimulation. The Fig.3.12 shows the threshold of the retinal cell in this experiment is the stimulating current while MPA is under 400mW IR illumination and 400mW IR illumination on the MPA can induce about 42uA photocurrent.

Therefore the threshold of the off α ganglion cell in this experiment is about 42uA. Further more, we carry on the same experiment with the on α ganglion cell. We can also record the cell response to verify the feasibility of the on α ganglion cell and the threshold of the on α ganglion cell is about 18uA. These experiment results can ensure that the silicon retina can

trigger the retina in the sub-retinal prosthesis. As a preliminary in vitro experiment of the silicon retina, the feasibility of the sub-retinal silicon retina is verified and the experiment data can provide some useful information to the future work.

Fig. 3.6 the ganglion cell used in this in vitro experiment. White bar represent 100um and the size of this ganglion cell is about 1.2mm. This picture is reprinted from NTHU Y.T. Yang.

Fig. 3.7. The extra-cellular signal recorded under varying light intensity of IR stimulus. The IR intensity vary from 100mW to 2000mW.

Fig. 3.8. The extra-cellular signal recorded under varying light intensity of IR stimulus. The IR intensity vary from 100mW to 2000mW. Each block is the zooming figure of Fig. 3.7. The spike with latency represents the cell’s response to the silicon retina’s trigger.

Fig. 3.9 The comparison between electrical stimulation and visible light stimulation. With the same experiment environment, including chip, retina, probe position and temperature, two responses from two different stimulations are recorded individually. Left block shows the signal which recorded while silicon retina was triggered by IR irradiation. Right block shows the ganglion cell response while the visible light triggers the retina. Despite the artifact, the latency response of both signals has some in common.

Fig. 3.10 The signal recorded under varying light intensity of IR stimulus with no retina on the chip. The IR intensity vary from 100mW to 2000mW.

Fig. 3.11 The comparison between cell response and silicon retinal artifact. With the same experiment environment, including chip, IR stimulation, probe position and temperature, two signals are recorded individually. Left block shows the signal which the retina is still attach on the silicon retina. Right block shows the signal while the retina is removed from the chip.

Under the same IR stimulation, the right block show some artifacts which is induced by the silicon retina. The same artifact is also shown in left block, but the spike with latency is clearly shows the feasibility of the silicon retina which can generate effective stimulation under IR stimulation. This experiment result can ensure that the silicon retina can trigger the retina in the sub-retinal prosthesis.

Fig. 3.12 the threshold estimation of the in vitro experiment. As we can see in this figure, the cell response is recorded occasionally. This means the threshold of the retinal cell is about this light intensity (or stimulating current).

3.4 DISCUSSIONS

As we can see in Fig. 3.7 – 3.11 , the retina cell is triggered by the MPA silicon retina is verified and the feasibility silicon retinal chip for sub-retinal prosthesis is also verified. We also carry on the same experiment with the on α ganglion cell and the threshold is about 300mW IR irradiation which means the stimulating current is about 18uA. In compare with the experiment in [48], the Fig. 3.13 and table II shows the comparison results and the experiment environment of the experiment mentioned in [48]. As we can see in the table II, both experiment animal are rabbits and the stimulation site are the same (PR side). According to Humayun’s report in [48], the electrode size has significant effect on the stimulus threshold.

The smaller electrode size has higher stimulation threshold. According to the table, the threshold measured in our in vitro experiment is very reasonable and the threshold is slightly higher due to the small electrode used in this experiment. But in the real application of the retina prosthesis, the IR stimulus with over 400mW is too large to destroy some eye tissue and the safety limit of the IR irradiation which applied in eye relative prosthesis is 10 sun light intensity which is about 100~200mW. Therefore the next generation silicon retina must solve this solar cell power issue.

Fig. 3.13 The experiment schematic of the in vitro experiment which mentioned in M.

Humayun in 2006.[48] Stimulus and recording electrodes were positioned to record retinal ganglion cell action potentials from axons projecting from the stimulus site. Stimulating electrodes were platinum disks, either 25 or 125 _m in diameter. Two tungsten microwires were used as a differential recording electrode pair. (a) Top view of the isolated retina chamber. (b) Schematic cross section of recording chamber. Retinal ganglion cell (RGC).

Both stimulating and recording electrodes were positioned using manual micromanipulators.

Since the stimulating electrodes could not penetrate the retina, the location of the stimulating electrodes (GC or PR) determined which side of the retina was placed face up in the chamber.

Recording electrodes were positioned first. When stimulating PR, the PR side faced up and the recording electrodes penetrated through the retina to the GC nerve fiber layer. When stimulating GC, the GC side faced up and the recording electrodes were inserted slightly into the GC nerve fiber layer.

TABLE II. the comparison of this work and M.Humayun in 2006 [48]

This work Mark S. Humayun,2006 [48]

Animal New Zealand Rabbits Dutch Belted Rabbits Target cell OFF α GC ON α GC GC (type not mentioned) Stimulating type current level current pulse

Stimulating site PR PR PR PR

Electrode size 10x10um² square electrode 25um diameter circle electrode

125um diameter circle electrode Threshold current 42uA 20uA 17.1±5uA 6.7±4uA

CHAPTER 4 Design and Analysis of Retinal Chip

4.1 CLOCK GENERATOR AND POWER CONTROL UNIT

Persistence of vision is the phenomenon of the eye by which even nanoseconds of exposure to an image result in milliseconds of reaction (sight) from the retina to the optic nerves. This is because persistence of vision depends on chemical transmission of nerve responses, and this biochemical hysteresis is much slower than the light transmission. A typical explanation of persistence of vision went something like this: when the human eye is presented with a rapid succession of slightly different images, there is a brief period during which each image, after its disappearance, persists upon the retina, allowing that image to blend smoothly with the next image. [47] Therefore the retinal chip needs not to stimulate the retinal cell continuously but stimulate the retinal cell with a more discrete method. As we know that the maximum frame rate which can be distinguished by human is about 60Hz, namely, we can stimulate the retinal cell every 16ms or less and the retina still consider the discrete stimulus as a continuous stimulus.

By applying the technique of divisional power supply architecture to exploit the characteristic mentioned above, a three times output current could be achieved. The details are in the followings. We divide the pixel array into four blocks whole outputs are controlled by four control signals generated by power control unit. The blocks will be activated in turn to send out their stimulating signals. The time interval between neighboring activation of the same block must be much smaller than biochemical hysteresis of the cell. Only one of the blocks is activated at the same time and the power from whole chip, which is supplied by solar cells, is provide to that block to increase the output stimulating current. In contrast to the conventional MPA design, the output current is much greater and the discrete stimulus won’t cause any misinterpret in the retina but still provide continuous signals to the brain.

The architecture of the chip is shown in Fig. 4.1. The block diagram of the power control unit is marked in the upper box of Fig. 4.1. The whole power control unit is only power supplied by on-chip solar cells. There are three main components in the unit, an oscillator, two frequency dividers, and combinational logic part. The schematics of the clock generator and frequency divider in power control unit are shown in Fig. 4.2. A ring oscillator is used to generate the reference clock. This structure is the same as the one in [50] which only on chip solar cell is acquired. The cross section view of the clock generator is shown in Fig.4.4. The on-chip solar cell system is the only supply of the clock generator and the NMOS is isolated to avoid the parasitic BJT causing the leakage current which result in the clock generator out of function. In order to generate the quadrature phase signal, a D flip-flop frequency divider shown in Fig. 4.4 is chosen to provide two synchronous signals with the half and quarter clock frequency. Two frequency dividers are required in this design because the clock signal from the clock generator is not in a square waveform. A clear square waveform with relatively short

rise time and fall time is required for the combinational logic circuit. Therefore we use two frequency dividers to provide two synchronous square-wave signals with the half and quarter clock frequency. We choose the conventional static cross-coupled NAND structure to implement the D flip flop of the frequency divider as shown in Fig. 4.5. The operating frequency of the frequency divider is about 1KHz to 10KHz which is much lower than the radio-frequency application. Therefore the modern dynamic D flip flop design which will result in charge leakage problem is not suitable for this work. The static design can operate in low voltage supply and also has lower power consumption. In the last stage, the combinational logic circuits consisted of NAND and INV gates synthesizes the final quadrature control signals. The logic gates mention above are powered individually by on-chip solar cell which each gate has six solar cells as the power supply. The schematic of the combinational logic circuit is shown in Fig. 4.6. The circuit schematic of the NAND gates and INV gates are shown in Fig. 4.7. The sizes of the MOSFETs in each block are shown in table III.

In the design of the retinal chip with on-chip solar cell supply, the total power generated by the solar cell is in proportion to the area of the solar cell. Therefore some layout area on the retinal chip is occupied by the power control unit. In other words, the power generated by solar cell in this work will smaller than conventional design because some layout area is occupied by control unit. The power saved by divisional power supply system is much higher than the power generated by the solar cell with same area of power control unit. The simulation results also show the output simulating current is about 3 times of the conventional one.

Figure 4.1. The block diagram of the retinal chip. The control circuit which provides quadrature phase control signals to pixel array is composed of pulse generator, frequency dividers, and combinational logic.

Figure 4.2. The schematic of ring oscillator provides the reference clock signal to control unit.

Dx1 Dx2 Dx13

M1 M2 M13

Figure 4.4. part of the cross section of the clock generator. We use the deep-n-well to isolate to NMOS from the leakage current of parasitic BJT.

Figure 4.4. The schematic of the frequency divider.

.

D N1 N2

N3 N4

I1 N5

I2 I3

Figure 4.5. The schematic of the frequency divider.

CK

D Q

input

output

Figure 4.6. The schematic of the combinational logic circuit.

Figure 4.7. The circuit schematic of the NAND gates and INV gates.

TABLE III. sizes of the MOSFETs of each block

clock generator D flip-flop combinational logic

NMOS (μm) PMOS/NMOS (μm /μm) PMOS/NMOS (μm /μm)

4.2 IMPLANTABLE RETINAL CHIP

The architecture of the chip is shown in Fig. 4.1. The power control unit generates quadrature phase pulse signals to activate the four blocks in the pixel array in turn. The whole chip is only power supplied by on chip solar cells to prevent the external wiring from hurting the eye ball. The quadrature signals mentioned previously control the activation of the blocks in the pixel array as shown in lower box of Fig. 4.1. According to the report of the artificial retina prostheses [2], we chose 16 pixels resolution in this chip. There are 16 pixels in the array which is divided into four blocks, and thus each block contains four pixels. The pixel circuitry is shown in Fig. 4.8. Diodes D2, which have 306 solar cells (photodiodes) in parallel connection, as the solar cell are the power supply of a single pixel. All the solar cells in the pixel array are connected in parallel to provide the global power supply to the whole array.

Photodiodes D1, which have 6 photodiodes in parallel connection, are the photo-sensor for each pixel whose photocurrent is amplified 200 times through current mirror for stimulating output. In order to provide large output stimulating current, we roughly estimate that each pixel should has at least 300 solar cells to produce at least 500nA output stimulating current under 1200lux illumination. In the prototypical design of the pixel circuit, there has only one photo-sensing diode. But that will induce another problem that the one stage current mirror, which is shown in Fig. 4.8 as M1 and M2, can’t afford more than 1000 times amplification and the mirror ratio will has mismatch. Therefore the amount of the photo-sensor photodiode D1 is chosen as 6 to keep the amplification ratio of current mirror in 200. The amount of diodes in D2 is chosen as 306 because there will be quiescent current in 6 diodes of D1 under the light illumination. 306 solar cells in D2 can ensure the photo-current induced by D1 be amplified 200 times. The output stimulating current is controlled by the M3 with the control signal from power control unit. The amplified current will be sent out for stimulating the retina tissue via the in-pixel electrode if the block is activated. When one block is activated, four pixels in this block receive the power from the parallel solar cells of all the blocks while the pixels in the other blocks consume almost no power because their output paths are cut off. Therefore, the output stimulating current will be approximately three times of the conventional MPA design.

In this work, about five thousand solar cells are connected parallel as the on-chip solar cell power supply. Four blocks are activated in turn that power wasted in refractory period is saved to be added to another blocks for more effective stimulation. The simulation results also show the power efficiency is elevated by the divisional power supply system.

Figure 4.8. The schematic of pixel circuit and array connection. The output electrode is controlled by the NMOS switch with the control signal from power control unit.

4.3 SIMULATION RESULTS AND LAYOUT DESCRIPTION

4.3.1 Simulation results

The simulation model of each photodiode in the circuit is established based on the measurement results shown in the previous section. The simulation model is shown in Fig. 4.9.

E1 is the voltage-control voltage source and F1 is the current-control current source, which present two kinds of characteristic of the photodiode: current source and passive component.

The photodiode will operate in 4th quadrant of I-V characteristic curve as a solar cell power supply while illumination, and operate in 3^rd quadrant as a photo-sensor. The magnitude of the photocurrent Iph in the model is chosen according to the measurement results of the testkey.

The photocurrent is assumed to be proportional to the area of the diode. All simulation results are based on device models of 0.18μm CMOS technology. We simulate two different ambient illuminations of 2.04mW/cm² light intensity and 3.6mW/cm² light intensity.

Photocurrent of D1 and D2 in Fig. 3 are 0.5nA and 1nA per each photodiode under two different ambient illuminations. The transient simulation results of the ring oscillator, two synchronous square-wave signals with the half and quarter clock frequency and control signals under different ambient illumination are shown in Fig. 4.10, 4.11, 4.12 and 4.14. The

simulation results of the output stimulating current are shown in Fig. 4.14 and 4.15. The time interval between neighboring activation is less than 3ms and output stimulating current is about

simulation results of the output stimulating current are shown in Fig. 4.14 and 4.15. The time interval between neighboring activation is less than 3ms and output stimulating current is about