Analog Neuromorphic Systems for Implantable Neural Interfaces

With the increasing popularity of brain-machine interfaces for a variety of motor disorders and fundamental neuroscientific experiments, there is also a growth in new sensors with increasing number of parallel channels to sense brain activity. Hence, there is an urgent need to manage the exploding datarate and power dissipation of these devices. Analog neuromorphic computing promises to be a natural candidate for bringing intelligence to implantable neural interfaces, thus only needing to transmit sparse information gleaned from the raw neural recordings. Neuromorphic event-driven sampling can reduce datarate at the source producing spikes that can natively be processed by low-power, sparsely activated spiking neural networks (SNN), that detect events of interest such as epileptic seizures or occurrence of biological action potentials (AP). This work compares popular neuromorphic analog-spike converters (ASC) in terms of their robustness to interference and shows delta modulation to be superior to integrate and fire modes of spike generation. Further, a new type of delta modulation inspired by high threshold bursting neurons in the hippocampus is shown to extend the tradeoff between compression ratio and data fidelity for conventional delta modulation. Measured results from fabricated integrated circuits show excellent data compression rates far exceeding those of conventional analog digital converter with Nyquist sampling. Further, mixed-signal, clockless SNNs interfaced with this bursting ASC are shown to be able to classify the generated spikes to detect AP events with high accuracy while dissipating scant energy.