46Gop/s of neural network processing in a 3.9 x 3.9mm package

Alif E1C npr mcu

Dubbed E1C, the family all get a single 160MHz Arm Cortex-M55 CPU with Helium vector processing extensions, then some also get the 46Gop/s via a 128MACs/cycle Arm Ethos-U55 NPU (neural processing unit) which includes on-the-fly weight decompression. Memory options are up to 2Mbyte of tightly-coupled zero wait-state SRAM and up to 2Mbyte of non-volatile MRAM. Fast OctalSPI allows external memory expansion.

“E1C shares the same architecture as the Balletto family, allowing for software re-use and migration of applications from one device to another,” according to Alif.

Peripheral options include dual 12bit SAR ADCs, a 24bit ΣΔ ADC, a 12bit DAC and a voltage reference.

Interfaces include MIPI DSI for external colour displays, USB 2.0, SDIO, dual CAN FD and I3C.

For energy saving there is a choice of four power modes including a 700nA stop mode.

“Manufacturers can bring ML [machine learning] capabilities to products such as wearable devices,” said the company. “Capabilities are optimised for local workloads such as object recognition, speech recognition, sensor fusion and adaptive audio noise cancellation.”

A hardware root-of-trust secure enclave with secure boot, cryptographic accelerators and certificate management is provided for key generation and storage.

Alongside the 90 bump WLCSP is a 120 bump FBGA for maximum I-O on four-layer PCBs, and a 64 lead TQFP for better temperature variation tolerance.

DK-E1 is the associated development kit.

Source

      Guidantech
      Logo
      Shopping cart