What’s next for Silicon Labs, the IoT and AI

The acquisition by TI is subject to shareholder votes and regulatory approvals but is expected to close early to mid-2027. Undoubtedly, the deal would bring benefits in terms of manufacturing. It is in step with the broader semiconductor industry which is, generally, moving to a foundry model, said Sabolcik, (Intel being the obvious exception).

“If you look at TI’s manufacturing capabilities, they are world-class,” said Sabolcik. “TI owning their foundries is unique, especially for analogue technologies. And since you’re not chasing those bleeding edge nodes with a lot of analogue, I think that’s where having your own manufacturing can make a lot of sense,” he continued.

Over the last 15-20 year, the world has evolved and today customers want more than just a chip today, Sabolcik observed. They want a microcontroller with wireless connectivity, the analogue and all of the software to control the peripherals, together with more ecosystem enablement and problem solving.

Wi-Fi is ‘having a moment’ in IoT, said Sabolcik. It is ubiquitious and needs little infrastructure to operate. For some devices, such as sensors which need high bandwidth capabilities, Wi-Fi is an attractive option, but for those looking for a robust, highly reliable, highly secure technology, there is still a place for PoE or EtherCAT, he said. He expects Wi-Fi 6 to coexist with other technologies. “PoE has the advantage of being hard-wired, you can do power over it for a hardened network,” he reasoned. Cabling has to be configured and takes up space, but in same cases it is the only solution. For monitoring a fixed conveyor, a wired option is an obvious choice but when a warehouse robot is moving parts bins and bringing them to another station, then the system cannot be wired.

He also thinks AI starts to become interesting when looking at the business model used to extract value from the data collected. For some applications, such as home security door or window sensors, rather than pushing a continuous stream of audio or video to the cloud via a ZigBee network, data can be analysed at the edge. This local processing option, based on M-class MCUs rather than large DSPs or GPUs, becomes attractive because it is fast and users want the edge devices to be operating even without cloud connectivity.

The company has recognised the advantages of processing data at the edge. Local processing is available with its Series 2 devices and Silicon Labs is now ramping up its Series 3 devices which have AI accelerators. One example is the BG24 series, with Matrix Vector Processing (MVP), a dedicated hardware accelerator designed to accelerate machine learning inferences and vector/matrix calculations in TinyML IoT applications. It accelerates processing speeds and reduces power consumption in SoCs by offloading task from the main Arm Cortex-M processor.

TI buys Silicon Labs