An analog cell library useful for artificial neural networks

The development of an analog cell library that is compatible with a standard digital CMOS technology is described. Several analog cells useful for the design of artificial neural networks (ANNs) have been designed. The initial efforts in developing an analog cell library utilized a bulk CMOS 3-μm double-poly single-metal p-well technology supported by MOSIS. The authors designed and laid out an amplifier, a resistor, a switch, a 12-b capacitance array, current mirrors, and several digital cells needed to implement a Hamming-type ANN. Based on this experience, the authors have started developing an ANN cell library that is based on scalable CMOS 2-μm double-poly double-metal p-well technology supported by MOSIS. They have designed the following circuits: a transconductance amplifier, a wide-range transconductance amplifier, a follower-integrator circuit, a wide-range Gilbert multiplier, and a current mirror. HSPICE simulations that have verified the functionality of the circuits and circuit descriptions are presented