ICY Tech Sets Out MRAM-Based AI Inference Roadmap for Rack-Scale Systems

ICY Tech has introduced uHBM and uLPU architectures for AI inference, according to DigiTimes. The company is positioning persistent MRAM-based computing across a roadmap that spans compute-memory dies, chips, modules, 2U trays, and rack-scale systems. The approach reflects efforts to address AI-inference workloads with closer integration between compute and persistent memory. If brought to commercial scale, such architectures could broaden the range of memory technologies considered for inference infrastructure. The supplied source does not provide technical specifications, manufacturing partners, customer commitments, commercial shipment dates, or performance comparisons. The announcement is therefore a technology-roadmap signal rather than evidence of material near-term displacement of HBM or conventional DRAM.