The EFLX4K AI eFPGA core is optimized for AI. In traditional FPGA the MACs are meant for signal processing and are overkill for AI. The EFLX4K AI core has 8-bit MACs (which can be configured as 16x16, 8x16 or 16x8). And the ratio of MACs to LUTs is shifted to more MACs. The result is >10x the MACs in the same silicon area as the existing EFLX4K DSP core. The EFLX4K AI has >10 times the GigaMACs/second per square millimeter of any existing FPGA or eFPGA. Available in 6-8 months in any CMOS node. The EFLX4K AI core can be arrayed to make larger MAC arrays or mixed with other EFLX cores, up to at least 7x7.
Block Diagram of the AI-Optimized eFPGA IP Core