Two-Dimensional Multiply-Accumulator for Classification of Neural Signals

Automatic spike detection and classification have been used for a neuroelectronic interface to reduce data amount or even to interact with neurons in a closed loop. While conventional neuroelectronic interfaces employ voltage-mode circuits to amplify neural signals and convert the signals into binary data, the dynamic range and signal-to-noise ratio of these circuits are directly limited by the supply voltage. To release this constraint, this paper proposes an analog-to-time converter (ATC), which uses positive feedback to convert analog neural signals into a sequence of pulse trains. Custom-designed digital circuits, including two types of time-to-digital converters (TDCs) and a 2-D multiply-accumulator (2-D-MAC), are further proposed for processing such time-mode signals. The ATC is implemented with the standard 0.35-μm CMOS technology and proved able to convert analog voltages into pulse-width-modulated signals with a resolution of 6 bits. The TDCs and 2-D-MAC are realized in FPGA and compared to the standard digital IPs. The comparison indicates that the TDC based on dual counters minimizes area consumption and the other based on delayed clocks minimizes power consumption. The 2-D-MAC further facilitates parallel computation of partial products and allows data to be classified without summing up all partial products. Finally, the application of the proposed time-mode system is demonstrated as classifying neuronal spikes.