Good news from AMD’s Chief Technology Officer, Mark Papermaster. Moore’s Law is not over. CPUs and GPUs will continue to get better in the near future. But he also has bad news. The ever-increasing cost of getting everything on track is driving the need for innovative solutions such as chiplet designs.
Moore’s Law, of course, is the observation that transistor density in integrated circuits doubles every two years. It was put forward by Intel co-founder Gordon Moore in his 1965, initially plotting an annual cadence of density doubling until Moore revised the schedule to his biennial in 1975. was doing.
speak in Summit in Las Vegas (opens in new tab)Papermaster explained that Moore’s Law is still on track, but chip technology is becoming increasingly complex.
“And you’ve heard many times that Moore’s Law is slowing down. Moore’s Law is dead,” says Papermaster. But according to AMD tech leaders, that’s not really true.
“It’s not that there aren’t exciting new transistor technologies. In fact, we can see the next exciting new transistor technology. As long as we can actually plan for these things, it’s about six to eight years, and it’s very, very “The advances we’re trying to make to keep improving transistor technology are obvious, but they’re more expensive.”
Papermaster said the difference now is that while transistor density used to double in the same year, the cost for a given chip size was roughly the same, but the cost per area of silicon has increased over a continuum. It is increasing for each production node. A computer of a certain size his chips have become much more expensive.
Papermaster says AMD foresaw it. This was a major driving force behind his move to chiplet designs for CPUs a few years back, and earlier this year for his GPUs with RDNA 3. If chiplets are part of the solution to rising silicon production prices, then combining old-school his CPUs and GPUs with specialized accelerators will be more important than ever.
“You have to use Accelerator,” says Papermaster. at our pace. “
In other words, while transistor density will continue to rise in line with Moore’s Law, chips will become more expensive, forcing companies like AMD to use chiplets to boost yields and keep costs from rising. It has been. Specialized circuits also tend to be much more compact than general-purpose CPU and GPU blocks, allowing increased performance without the price penalties associated with traditional pure CPU or GPU designs.
In recent years, the imminent demise of Moore’s Law has been widely reported. Part of the reason lies in the conspicuous struggles of the former king of chip-making technology, Intel. But Intel is not the only chip maker. While certainly behind the curve, his TSMC at the cutting edge has kept Moore’s Law on track.
Of course, TSMC makes AMD’s CPUs and GPUs, not to mention NVIDIA’s latest GPUs and Apple’s M1 chip. Currently, TSMC’s 5nm node seen in AMD’s CPUs and GPUs is about a generation ahead of Intel’s 10nm node which was recently rebranded as Intel 7.
TSMC has also just started 3nm production, and devices using TSMC 3nm silicon are expected to hit the market in the first half of 2023, mostly on Apple Mac computers running the 3nm derivative of the M1 and M2 chips. will take the form of TSMC expects his 2nm to start production in 2025. Moore’s Law looks sound enough at this point. It’s just a little more maintainable.
