We have recently been fielding a few conversations about the future of analog chip companies. Amidst all the recent excitement around AI and all the digital semis that requires, many people forget about the analog side of the industry. Optical is the exception, as these are an increasingly important part of the AI networking stack, but what about all the micro-controllers (MCUs), digital/analog converters, sensors and actuators that the industry produces in immense quantities. How do those companies best position themselves for the future?
This is actually a recurring theme here. We published on this topic a year ago, and probably the year before that as well. We used to work at an analog company, so apparently our social circle is replete with analog people wondering what their future holds. And this is a shame – analog chips are both commercial and technical marvels. Dig into the capabilities of some of the products coming on the market from the analog companies large and small and you can start to question the laws of physics.
Despite that, with the biggest shift in compute in a decade underway in the data center, the big analog chip companies are rightfully worrying that they may be left behind. Is there a way for analog companies to participate in the AI Bubble Boom?
Unfortunately, there are no easy answers. Designing analog chips and digital chips are very distinct skill sets, with little overlap between the two. There is almost an entirely different mindset with the two sides looking to solve very different sets of problems. We sometimes hear the comparison between chefs and bakers – they use a lot of the same tools, and they are both trying to feed people, but the way they go about their respective tasks are radically different.
In practice the analog chip companies have little to no digital design talent. This means they now face a choice as to how to grow their digital capabilities. They could start a digital team and try to grow it organically. This has the advantage of being fairly low-risk and easy to manage, but it will take a long time. Moreover, the new team faces the massive obstacle of starting life without a clear commercial purpose and will perennially be at risk of being drained of resources by the analog teams which bring in all the revenues. Hard to see why a digital engineer would want to go work in that kind of environment when there are plenty of open jobs at Nvidia.
The other choice is to bring in outside IP via acquisition or licensing. The licensing path makes a lot of sense around specific projects, but the devil will be in the details. This leaves acquisition as probably the best path forward. Of course, M&A carries significant integration risk – see above where the newcomers soon become the internal team bowled over by their analog peers. That being said, this path actually looks fairly promising. For starters, there a lot of potential targets, 60+ at last count. There are literally dozens of start-ups trying to design chips for “AI at the edge”, and in many cases a key feature of those products is some form of interaction with analog parts. Give these teams some time and room for experimentation.
Our suggestion for large analog companies is to take a look at all the circuit boards where their chips currently sit. Think through where some digital or AI functionality could add value. This will show up in surprising places. For instance, there is a surprising amount of demand for mid-range MCUs with just a touch of AI circuitry for use in automotive applications. In many situations, a cheap MCU with “good enough” digital compute has enough horsepower to deliver on tasks that others would want to do with an expensive all digital accelerator.
The other step we would recommend is to hire a few software architects. In some senses, the biggest difference between sales of analog and digital chips is the role that software plays. For digital chips software is all that matters, while in analog software does not matter at all. The analog companies should hire a small time of expensive software architects to help these companies think through what software they can use, to search out the optimal boundary between increasing chip functionality and decreasing software size. Our guess is that these teams will come up with some surprisingly powerful performance with surprisingly small chips.
Leave a Reply