AMD architect scales challenges with 14nm/16nm process

There's still a race for performance, he argued, pointing to emerging requirements to support 4K video. "Periodically computers get fast enough to run existing apps or even too fast and then new apps emerge and nothing is fast enough again—that happens over and over," he said.

The challenges building the processors are as great—and solvable—as ever, but the dynamics of those changes have shifted, he said. "In the good old days, a custom process and processors were built together."

Now AMD and all but a few such as Intel are fabless designers. They use standard process technologies, but they get an early look at what's coming down the pike.

Keller considers largely solved the issues dealing with multiple-patterning lithography required for the 14nm/16nm processes in which his team is designing the first K12 chips. Just adding on more metal layers is no longer an option, but "we have other proprietary tricks," he said.

EE Times' Rick Merritt sits down with AMD microprocessor architect Jim Keller to discuss about the challenges of building processors on the basis of multiple-patterning lithography, FinFET transistor count, and Moore's Law.

Jim Keller is not buying the stories that Moore's Law is dead or dying. In his many years designing microprocessors, that prediction is just one more evergreen, he said.

I only had a few minutes with Keller yesterday. But it was much more time than I had in the years he spent at Apple helping design SoCs for the notoriously secretive company's iPhones and iPads.

"I worked behind three lock doors in those days," he quipped at a Monday event where he talked about his current work designing x86 and ARM cores and SoCs for Advanced Micro Devices.

It's back to the future for Keller, who helped AMD get a leg up on Intel back in the day with the K7 and K8, its first 64bit x86 CPUs. Now AMD hopes its returning star will help it differentiate the K12, another 64bitter, from a pack of ARM architecture licencees that includes Applied Micro, Broadcom, Cavium, Nvidia, and Qualcomm.

It's just like the old days with plenty of competition in processors but "the players have changed a bit," Keller said.

There's still a race for performance, he argued, pointing to emerging requirements to support 4K video. "Periodically computers get fast enough to run existing apps or even too fast and then new apps emerge and nothing is fast enough again—that happens over and over," he said.

The challenges building the processors are as great—and solvable—as ever, but the dynamics of those changes have shifted, he said. "In the good old days, a custom process and processors were built together."

Now AMD and all but a few such as Intel are fabless designers. They use standard process technologies, but they get an early look at what's coming down the pike.

Keller considers largely solved the issues dealing with multiple-patterning lithography required for the 14nm/16nm processes in which his team is designing the first K12 chips. Just adding on more metal layers is no longer an option, but "we have other proprietary tricks," he said.

For the 14nm/16nm process, his team had to scale challenges with the vertical FinFET transistors it uses. He notes of all the parts of his team, the design for manufacturing and test groups have grown the most over the years.

The 40-person circuit design teams once needed for a major microprocessor have shrunk. But more engineers are needed to write RTL and verification code. "About two-thirds of the team writes software every day for a living," Keller said.

One piece of welcome news is that the architects who still do their work with magic markers on white boards "sometimes write verifiable specs which was a dream 20 years ago," he said.

Overall, Keller is still upbeat. Looking more fit and trim than ever, he's an engineering manager up for a hefty challenge. He's even had some early looks into the 10nm node, and he's opaquely optimistic about it. "It has more transistors," he deadpans.

- Rick Merritt
??EE Times