A 0.7 V low-power fully programmable Gaussian function generator for brain-inspired Gaussian correlation associative memory

作者：Li Fei^*; Chang Chip Hong; Basu Arindam; Siek Liter

来源：Neurocomputing, 2014, 138: 69-77.

DOI：10.1016/j.neucom.2013.02.060

• Brain-inspired
• Gaussian function generator
• Gaussian correlation associative memory
• Analog circuits
• Subthreshold

An analog current mode programmable Gaussian function generator (GFG) suitable to be used as the nonlinear transfer function in the VLSI implementation of a brain-inspired associative memory model is proposed. It mainly consists of a current absoluter, a translinear loop, a MOS resistor and a PMOS transistor with most of the transistors biased in the subthreshold region for very low voltage and low power operations. The mean, gain, and standard deviation of the Gaussian function are independently and readily programmable. Post-layout simulation results show that the circuit operates well at 0.7 V supply voltage and consumes a maximum power of only 0.6 mu W in 0.18 mu m CMOS technology. Overall, it is at least 62% more efficient in terms of millions of multiply accumulates per second per milliwatt when compared to other analog GFG circuits or DSP implementations.

• 出版日期2014-8-22
• 单位南阳理工学院