My most recent post took a look at some of the major technology stories for 2023. Needless to say, a lot of attention is centered on generative AI. I ended by saying “Let me go out on a limb with a forecast for 2024: AI will be on everyone’s 2024 trends lists and, come next December, on everyone’s end of year roundup.”
We won’t know about the year-end 2024 roundups for another 11 months, but in terms of trend lists, it sure looks like I called it – at least with respect to Nitin Dahad’s January 2nd piece on embedded.com. He did so through the prism of the embedded technologies that will be at the
heart of the products will be on offer at CES 2024, which just finished up (and which I’ll be posting about next time around). Dahad starts out by noting that:
…after the industry ‘hype’ of 2023 around generative AI, [after CES 2024]consumers will begin to understand what it means for them in their everyday lives. Almost every industry vertical will see more connected embedded devices with even more smartness or intelligence at the edge…That means more machine learning (ML) in more and more constrained devices, in the sensors, whether it is for the internet of things (IoT), for industrial automation, for autonomous mobility and software-defined vehicles (SDVs), or for health and wearable devices.
There are three “embedded world” trends that Dahad believes will help bring this about.
First up, he sees improvements in edge intelligence. He focuses his argument on a recent research paper from Apple which addresses the problem of running the large language models (LLMs) essential to AI on devices which are limited when it comes to memory. The Apple solution to “efficiently running LLMs that exceed the available DRAM capacity by storing the model parameters on flash memory but bringing them on demand to DRAM.” Apple claims to have achieved notable success in running large models in constrained environments by reducing “data transfer by reusing previously activated neurons” and increasing “the size of data chunks read from flash memory.” Voila! Edge intelligence equipped for great machine learning and inference.
Dahad also predicts greater visibility when it comes to RISC-V adoption. One indication of the RISC-V market heating up is the founding of Quintauris, a company (backed by some pretty high-powered investors, including NXP Semiconductors and Qualcomm) that aims to become the go-to source for all things related to RISC-V based products. Dahad thinks that RISC-V International, the industry association” will be the keepers of instruction set architectures (ISA’s) specifications, and that Quintauris will serves as “a resource for developers needing read-made reference boards and systems for their own development.”
The final embedded technology trend Dahad lists is that the “chiplet business will start looking a bit like IP business.” What does this mean? Well, chiplets are “one of the answers to overcoming the challenges [such as chip scalability, modularity, and flexibility] of enabling the massive compute demands from today’s ML-intensive products, without having to build one monolithic chip at the most advanced (and expensive) process technology available.” Given the promise of chiplets, we should see a buildout of the chiplet business and a proliferation of tools and designs coming from a range of sources.
Again, what’s driving these trends is the growing demand for generative AI/machine learning in all sorts of apps and devices – many of which will have no doubt been on display (virtual and physical) at CES 2024.
Image source: Forbes