Loading...
Citrini Research states that AMD and Apple are advancing the replacement of DRAM with flash memory in AI products to reduce memory costs. AMD acquires MEXT to optimize flash performance, while Apple achieves device side optimization through LLM in a flash technology. In AI inference, the demand for KV cache has increased and HBM occupies 25% of DRAM capacity, while the cost of flash memory is only 1/55 of DRAM. Through controller optimization, NAND stacking, and unit mode adjustment, a capacity and bandwidth alternative solution can be provided for edge AI.