AMD Chips on Chips?
By Ed Sperling -- EDN, May 17, 2006
San Jose—Advanced Micro Devices said it will begin looking at streamlined processors with specific functions to add first outside of its powerful general-purpose processors, with the likelihood those external processors will become part of the main processor in coming years.
The fact that AMD is even considering adding additional processors into its bread-and-butter chips represents a major departure in thinking at one of the Big Two processor companies. But with complexity and the cost of development on the rise, not to mention a growing focus on power conservation, established chip companies are beginning to think what two years ago would have been heresy.
For one thing, adding special-purpose processors for the first time opens the door to other companies beside Intel and AMD, which were frozen out of the processor market by the prohibitive costs of developing bleeding-edge multi-purpose chips. Add to that the intense and costly battles between the two giants, along with the shrinking number of customers who can use those chips, and the efforts of other semiconductor companies in this space have largely evaporated.
Multiple special purpose chips working in conjunction with general processors—and eventually within them—changes that dynamic. If it comes to fruition—and there certainly are no guarantees—it offers third-party processor makers, embedded developers and IP companies a potential inroad into what until now has been a closed market.
“We will do this only when there is a clear return on investment,” said Charles Moore, senior fellow at AMD. “If you run an application, the question is whether you can run the inner guts on the CPU and offload other parts onto hardware that does it more efficiently.”
Moore said the move will require far tighter integration between hardware and enablement software. He said that if those are not in place, the results could be “dangerous.” But he said the payback can be significant if done right, and that AMD will develop application programming interfaces in the future, similar to what currently is being done with DirectX, where some systems now have a dedicated encoder.
“This will start externally,” he said. “If it moves to internal and external development, we will enable key partners to build stuff with us.”
Steve Leibson, strategic marketing manager at Tensilica, said that adding specific-function chips could greatly extend the battery life of a device. He noted that powering down the Windows operating system could double a PC’s battery. That would allow a specific-purpose decoder chip, for example, to run DVDs for far longer than they currently run on a PC.
“The display still pulls about as much power as the operating system,” Leibson said. “But this could still double the battery life. You can shut down the main processor and play back MPEG 4.”
The whole approach raises interesting questions about what exactly an integrated design manufacturer (IDM) will look like in the future. Max Baron, principal analyst and senior editor at the Microprocessor Report, said the real benefits to such an arrangement are power savings and time to market. He said adding an accelerator or additional devices into the chip means a user can power processors up and down as needed.
“The goal is to not add complexity to the core processor simply and instead add things around it,” Baron said. “That makes it faster time to market and adds less risk. The buses are already established and the caches are established. Advanced chips practically demand this. But if AMD does it, it also will mean interesting pressure is applied to others.”
Baron noted that the embedded world already is creating chips that resemble general purpose processors. He said adding specific processors into chips such as AMD’s multicore offerings would push the processor market toward the embedded world, as well, blurring the lines in both directions.
