Deep learning consideration with novel approach — Look-up-table based processing conjugated memory

This paper is the progressing study of our previous one
⁽¹⁾
on the IMPACT 2015 and 2017 as the MLCS (Memory Logic Conjugated System) that is applied with LUT processing. It has been researched about how to fit the MLCS for Deep Learning (DL). Nvidia
⁽²⁾
, Google
⁽³⁾
, Fujitsu
⁽⁴⁾
Intel-Altera, Intel-Nervana
⁽⁵⁾
and Renesas
⁽⁶⁾
recently said that 8bits processing keeps efficient and flexible execution for AI, especially in DL on which we already discussed in the previous study. In this paper is discussed by the actual MLCS circuit implemented in a FPGA for a DL approach. Perceptron method for DL is examined and resulted fairly fitting for DL. Our study also has done by LUT processing to apply to DL that is just memory operation comparing with normal calculation method. The conclusion reads the all memory system (MLCS) can be achieved dynamic reconfiguration, the highest IO band and lowest power consumption with high speed.