This Neural Network Runs Without a CPU or Operating System
Machine learning projects almost always have a software component. Even when dedicated AI hardware is involved, there is typically a processor running an operating system, loading a model into memory, and executing an inference engine. Zakriya Paracha wanted to see what would happen if all of that disappeared. His solution was a neural network that exists entirely as hardware inside an FPGA.
The project recognizes handwritten digits captured by a camera, but there is no CPU, no operating system, and no software involved once the FPGA has been programmed. Instead, the complete inference engine is synthesized into digital logic written in Verilog. Every multiplication, memory access, and decision is performed by dedicated circuits rather than instructions executed by a processor.
That makes this project very different from one that simply accelerates AI on programmable logic. The FPGA isn't acting as a coprocessor. It is the neural network.
The complete setup...
Copyright of this story solely belongs to www.hackster.io. To see the full text click HERE
