FlexLogix has implemented an FPGA core consisting of an array of MACs with smaller multipliers (8/16bit) than MACS optimised for DSP which claim to deliver more neural network processing per sq mm than DSP-optimised MACs The core is called the EFLX4K AI eFPGA. It claims to improve deep learning performance 10x AI needs more MACs/second, more ...

This story continues at AI-optimised eFPGA core

Or just read more coverage at Electronics Weekly