Title :
A spiking-neuron collective analog adder with scalable precision
Author :
Sung Sik Woo ; Sarpeshkar, Rahul
Author_Institution :
Dept. of Electr. Eng. & Comput. Sci., Massachusetts Inst. of Technol., Cambridge, MA, USA
Abstract :
Although neurons function with locally imprecise analog computational units, they can collectively interact to achieve computations with high complexity, low error rate, or high precision. In this work, we show how many moderately precise integrate-and-fire analog units interact via the mechanism of a spiking `carry´ to collectively add numbers to arbitrary precision. For example, in a proof-of-concept implementation that we implemented with chips built in a 0.5 μm CMOS process, we show that four 4-bit precise analog units can collectively interact to implement 16-bit-precise addition. Errors in the analog computation are minimized via novel time-based calibration techniques. Our work may lay a foundation for future collective analog computation that is arbitrarily precise just as current digital computation is today.
Keywords :
CMOS analogue integrated circuits; adders; neural nets; CMOS process; analog computational units; arbitrary precision; collective analog computation; integrate-and-fire analog units; neuron function; precision scalability; size 0.5 mum; spiking carry mechanism; spiking-neuron collective analog adder; time-based calibration technique; Adders; Calibration; Capacitors; Clocks; Delays; Switching circuits; Threshold voltage;
Conference_Titel :
Circuits and Systems (ISCAS), 2013 IEEE International Symposium on
Conference_Location :
Beijing
Print_ISBN :
978-1-4673-5760-9
DOI :
10.1109/ISCAS.2013.6572172