TinyEngine NPUs for MCUs
Texas Instruments expanded its MCU lineup to include a TinyEngine NPU aimed at efficient edge AI for cost‑sensitive real‑time and industrial applications—so you can run small neural nets on microcontrollers without a big bill. This directly targets embedded/IoT products that need on‑device inference. (x.com)
TI unveiled the MSPM0G5187 and the AM13Ex (AM13E230x) MCU families at embedded world in Nuremberg on March 10, 2026. (ti.com) TI’s product brief quantifies the accelerator’s impact as up to 90× lower latency and more than 120× lower energy per inference versus comparable MCUs without an accelerator. (ti.com) In the AM13E230x family TI rates the TinyEngine at 2.56 GOPS with support for 8-, 4- and 2‑bit precision and mixed‑precision modes. (eetimes.com) TI positioned the MSPM0G5187 as an Arm Cortex‑M0+ general‑purpose MCU and the AM13E230x as an Arm Cortex‑M33 real‑time MCU aimed at adaptive motor control and predictive‑maintenance applications. (eetimes.com) TI says the MSPM0G5187 is in mass production and listed on TI.com, while the AM13E23019 is in pre‑production with additional packages and memory variants slated through the end of 2026. (prnewswire.com) The company bundles CCStudio IDE with integrated generative‑AI features and CCStudio Edge AI Studio, which TI says ships with more than 60 prebuilt models and application examples for training and deploying networks on TinyEngine MCUs. (ti.com) TI highlighted AM13Ex motor‑control specifics — support for up to four simultaneous real‑time motor control loops and a trigonometric math accelerator claimed to be about 10× faster than a CORDIC implementation — and estimated multichip substitutions can reduce bill‑of‑materials cost by as much as 30% in those systems. (eetimes.com(ti.com.cn)
