Customizing RISC-V Cores to Accelerate Neural Networks - Jon Taylor, Codasip
AI and ML techniques are no longer just for the data center but are increasingly applied in IoT and at the edge. But that requires them to run cost-effectively with low power and on simple micro-controllers, not sophisticated dedicated processors. This talk will explain how the Codasip L31 core can be boosted with compact AI accelerator and custom instructions that improve the performance of a simple NN algorithm by >3x while halving the energy used. We will show how the code was profiled and custom instructions selected, before being simulated and implemented using Codasip Studio in only 200 lines of code.
