➀ Numem, a Sunnyvale startup, has developed MRAM-based chiplets designed for AI systems. Sampling is expected to start in late Q4'25.
➁ The chiplets offer up to 4TB/s memory throughput and support 4GB per stack package, making them scalable for demanding AI workloads.
➂ Numem's technology claims to significantly reduce standby power and deliver up to 4x faster performance than HBM while maintaining ultra-low power consumption.