A Resist-Protection-Oxide Transistor With Adaptable Low-Frequency Noise for Stochastic Neuromorphic Computation in VLSI

摘要

Noise is found to play a beneficial rather than harmful role for neural computation. For example, the sensory neurons exploit stochastic resonance to enhance their sensitivity. This finding has inspired several neuromorphic systems attempting to use noise for computation. Nevertheless, an adaptable noise source is essential for taking the most advantages of noise. This letter presents a resist-protection-oxide (RPO) transistor, which is a defect-rich transistor between the drain implant and the gate. The RPO defects enhance greatly the low-frequency noise of the transistor. The noise level is further adaptable over two decades by the drain voltage. Moreover, the transistor is fully compatible with the standard CMOS logic technology without requiring additional masks or process steps. All the features underpin the development of stochastic neuromorphic computation in integrated circuits.