Silicon Labs Announces BG24, MG24 TinyML 2.4GHz SoCs — and Boasts of a Fourfold Performance Gain
Company makes some bold performance gains for its new parts, which include native TensorFlow support.
Silicon Labs has announced the launch of a new system-on-chip range designed for edge AI projects — with the claim of a fourfold performance boost in one-sixth the energy, and with support for the Matter Internet of Things (IoT) connectivity standard.
"The BG24 and MG24 wireless SoCs represent an awesome combination of industry capabilities," boasted Matt Johnson, chief executive of Silicon Labs, at the launch of the new parts, "including broad wireless multi-protocol support, battery life, machine learning, and security for IoT edge applications."
The core specifications of the two families are identical: An Arm Cortex-M33 processor running at up to 78MHz with floating-point unit and digital signal processor (DSP) instructions, up to 256kB of RAM and 1,536kB of flash, and what Silicon Labs describes as "dedicated AI/ML [Artificial Intelligence/Machine Learning] accelerators […] showing up to a 4x improvement in performance along with up to a 6x improvement in energy efficiency." These, the company has confirmed, support TensorFlow workloads as well as use in tools from Edge Impulse and SensiML.
Alongside this, the parts also include an on-board 12/16-bit analog-to-digital converter (ADC) supporting up to one megasample per second (1Msps) at 12-bit resolution, two analog comparators, two digital to analog converters (VDACs), up to 32 general-purpose input/output (GPIO) pins, a 16-channel peripheral reflex system (PRS) and eight-channel direct memory access (DMA) controller, two I2C interfaces, two USART, and two EUSART buses, among other peripherals.
Where the two differ is in their respective radios: While both pack an on-board 2.4GHz radio, the BG24 range uses it to support Bluetooth 5.3 Low Energy with Bluetooth Mesh capabilities while the MG24 adds OpenThread, Zigbee, and Matter compatibility on top.
Silicon Labs has not yet confirmed pricing for the two parts, which launch in QFN40 and QFN48 package options, with shipping to "alpha customers" beginning to day and general availability scheduled for April this year. The company has pledged to launch the parts in module format in the second half of 2022.
More information is available on the BG24 and MG24 product pages.