The Flash Memory Summit, held in Santa Clara, California, August 6 to 8, covers non-volatile memory and solid-state drives. Weebit will provide a demo based on the latest SiOx ReRAM technology that runs inference tasks use spiking neural network (SNN) algorithms developed by CEA-Leti (see SiOx ReRAM selected for neuromorphic research).
Weebit is on record as producing ReRAM arrays using 28nm CMOS manufacturing process on 300mm wafers.
The ReRAM architecture implement synapses in a way which mimics the spiking of human biological synapse activity, Weebit said. This is different to the way the majority of neural network accelerators operate, which more simply implement gross synapse functionality.
It will also show increased parallel connectivity compared to conventional NN circuits. This will result in extremely energy efficient, fast and dense AI computation in the future, fitting many more types of applications, Weebit said.
“While our focus as a company, and thus our partnership with CEA-Leti, is on standard uses of non-volatile memory, we are not neglecting the huge potential of AI in the future. CEA-Leti is a leading research institute in the domain of neuromorphic computing,” said Coby Hanoch, CEO of Weebit Nano, in a statement.
Weebit did not disclose the complexity of the circuitry used in the demonstration but said it enables massively parallel, low-power and low-latency computation.
Related links and articles: