STMicroelectronics says edge AI growth hinges on in-memory computing shift
STM•Toolchain support for adoption
STMicroelectronics emphasized toolchain readiness as key to adoption, citing compiler and quantization support within ST Edge AI tools and the STM32 AI ecosystem.
Accelerator and commercialization path
The company highlighted an 18 nm FD-SOI digital in-memory accelerator shown at ISSCC 2023, delivering 40 to 310 TOPS/W at up to 4-bit precision.
It pointed to Neural-ART NPUs as the commercialization path, including STM32N6 as the first STM32 with built-in hardware AI acceleration.
Edge AI strategy centers on in-memory computing
STMicroelectronics management flagged edge AI as shifting from cloud training costs to inference economics, pushing compute closer to data sources.
Strategy centers on in-memory computing to cut energy from data movement, moving from near-memory designs toward SRAM and non-volatile in-memory architectures.
