The neural chip is a highly sensitive device detection and processing device. We faced some challenges along the way in the development of the hardware. We wanted to reduce the noise generated by the device by random firing of neurons, and to make the device as robust as possible. We initially took design inspiration from MEA60 device, for reading data from cortical neurons. This ran into the issue of increased noise due to high number of output electrodes. We decided to keep the number of electrodes to a minimum to reduce on noise. We also increased the surface area of the output electrodes and increased the number of input electrodes, by designing two input electrodes for the same signal, to overcome the issue of one of the electrodes breaking. With each iteration of our chip design we managed to reduce the noise generated by our system and achieved better and more accurate predictions.

From this first step, we decided that we needed to use vinyl masks for sputtering and had to increase the number of input electrodes to 6, although the number of input parameters was 3. We kept 2 electrodes for the same input signal.

Design 1 worked better than design 2, both in the quality of sputtering and neural connectivity.

This worked and in the next round of testing, we reported fewer than 10% cell death after a week of incubation.