AI inference requires huge amounts of memory, and a significant part of it contains coefficients and other data which is mostly read. It needs fast access to this memory. The huge computation requirements push it to use the smallest geometries where flash is not available. ReRAM (RRAM) is a very good option, as it can be embedded in the AI chip, eliminating large discrete NVM dies, and even replacing some of the SRAM. Since a ReRAM cell operates very similarly to a synapse in your brain, it is also ideal for neuromorphic computing and In Memory Compute (IMC), and many research institutes today use Weebit ReRAM as the basis for their research in this domain.