EE Awards / DeepMentor / CMO accelerator
CMO accelerator
Candidate for:EE Awards - IP/ Processor
CMO Accelerator, short for Cross Model Optimization, is a CGRA-based systolic-array NPU IP core delivering high INT8 TOPS performance for edge inference. Unlike fixed-function accelerators, it is AI model-driven: built on DeepMentor's patented Algorithm Hardening technology, protected by an issued US patent and a 12-patent portfolio, which converts each AI model's own computation graph directly into RTL, so the hardware is shaped around the specific model rather than a generic template. Its 16x16 PE systolic-array Kernel Engine with 64-lane SIMD is fed by a 4-channel, 6D-stride DMA engine and a 16-bank, 4MB Share Pool SRAM for efficient on-chip data reuse, and integrates directly with a RISC-V core to function as a true coprocessor rather than a bolt-on block. This methodology is validated in a peer-reviewed paper showing an 11.6x INT8 GEMM speedup from a related RISC-V NPU instruction extension — proof it scales from algorithm to silicon-ready hardware. Voting for CMO Accelerator means backing an AI model-driven edge accelerator that is already validated on real hardware and ready to license.
Malicious vote manipulation is expressly forbidden in this voting event. The organizers reserve the right to evaluate the fairness and accuracy of the voting results. AspenCore retains the authority to interpret the rules of this event.