Embedded Memories for Silicon-ln-Package: Optimization of Memory Subsystem from loT to Machine Learning

Traditional memory subsystem consisting of SRAM, DRAM and SSD/HDD has served the needs of electronics industry for decades. With the rapid increase in Graphics, loT and Machine Learning applications, several new memories have been innovated to optimize the memory hierarchy. Optane memory is deployed to close the performance/area gap between DRAM and SSD. Similarly, embedded DRAM inserted in products as L4 Cache to close the gap between SRAM and DRAM. Lately, MRAM and ReRAM are also brought to reality to target wide range ofapplications, covering embedded Non-Volatile Memory and Flash buffer at platform level. This talk will go through these innovate memories and how one needs to optimize the memory subsystem for the best application. There's no memory that fits all, but design and architecture opportunities exist for targeted applications.