A new memory designer, PieceMakers, has begun trading on the Taipei exchange with backing from DRAM maker Nanya. The company's central bet is that edge AI devices will not depend on HBM, the high-bandwidth memory used in data center accelerators, but will instead use custom memory fused directly to the processor.
PieceMakers' approach uses hybrid bonding to attach a DRAM stack directly to the processor die, avoiding the separate HBM packages typical of AI accelerators. The company believes inference workloads at the edge will move away from the HBM-centric model.
The Sept. 16 listing signals investor interest in alternatives to HBM for AI workloads. The source notes this is a bet on future edge AI memory architecture, though it does not provide details on performance targets or production timelines.