Researchers at GW have developed a novel, cost-effective, and energy-efficient residue number system (RNS) adder and multiplier that is based on integrated nanophotonics. The disclosed invention utilizes multiplication-accumulation computation (MAC) operations that can use one summand/multiplicand repetitively millions of times. In other words, aspects of the disclosed photonic RNS engine is well suited for deep learning application such as can be used in convolutional neural networks by transferring a given weighted matrix into switch states. The disclosed invention: can open the door of computing while switching; is better than any other optical RNS adder and multiplier in terms of the product speed, area and energy consumption. Further, the utilized broadband operating spectrum enables the device to support multiple channels with different wavelength, which allows multiple operations to run simultaneously.
The disclosed invention can be implemented as either a system, a method, or as a device as can be appreciated. The system or method or device can include various aspects as follows: (i) an RNS adder that can have optical switches to ensure strict nonblocking communication; (ii) a plurality of switches that can be placed on a set of parallel waveguides so as to allow light to propagate any possible path in the router by changing the control signal after; (iii) photodetectors at output ports that are capable of translating the result of an operation to capture a spatial position. In an embodiment, the RNS multiplier has a structure similar to the RNS adder as can be appreciated.
Fig. 1 – Aspects of the disclosed invention
Applications:
Advantages: