This is a reprint of our original article that ran in the EETimes.
Using proprietary processor architectures makes it hard to switch silicon vendors down the road, reducing your flexibility and increasing cost.
You’re ready for your new project. You know what features your customers want and what performance, power and cost targets you need to reach. You’ve even selected the best semiconductor solution to embed into your next-generation device.
Your next steps are to develop the product itself: the board, the mechanics, and of course lots of software development for the CPUs, GPU and application-specific DSPs that are included in the SoC.
After a year of development your product ships to market, and it does well. There’s another generation planned, and your silicon supplier has even developed a new SoC that is more powerful than what you were using before. You continue to add features to the product, and the software component of your device keeps growing.
But after a few product generations, your semiconductor supplier is increasing its prices, even though its devices aren’t as competitive as they used to be.
The time is right to switch suppliers and adopt another semiconductor solution.
Unfortunately, it turns out that a large portion of the software you’ve developed can’t be easily ported to another vendor’s silicon. They have different processor architectures on board, and it will take many man-decades to port, adapt and reoptimize all the software. You’re stuck between a rock and a hard place. You’re locked into your silicon vendor’s solution.
The cycle described above is all too common. Even sub-$5 SoCs have grown into complex multicore systems that have several software-programmable processors on board, each with its own software development tools and instruction sets. The SoC vendors understand this cycle very well, and they purposely try to lock you into their proprietary architectures.
Is there a remedy?
Luckily, there is.
Ensure that your SoC vendors use processors that are available on the open market as processor IP.
Case in point: ARM. Instead of selling ICs itself, ARM licenses its processor designs to the semiconductor manufacturers, who integrate it into their SoCs.
The net effect is that applications written for an ARM can be easily ported from silicon vendor A’s device to silicon vendor B’s, the same way the applications on your mobile phone run with equal ease on a processor from Qualcomm or Mediatek. That makes it easy for the system developer to switch vendors.
This is already standard practice for host CPUs, but in many other processor domains it has yet to arrive. Many semiconductor vendors still develop their own GPUs, DSPs, and communications, imaging and computer vision processors. Not because they can do better than what’s available on the open IP market, but to lock you in, and ultimately extract more dollars from your pocket.
Demand that your SoC vendor adopt a licensable architecture and you’ll reduce your risk and cost, as well as increase your future options.