Comms held Pentium 4 team together

Comms held Pentium 4 team together
By Terry Costlow, EE Times
November 1, 2000 (7:07 p.m. EST)
URL: http://www.eetimes.com/story/OEG20001101S0077

The PC industry will glimpse its future in the next few weeks, when Intel introduces the Pentium 4. As the first "blank paper" CPU design from Intel in quite some time and the preoccupation of hundreds of engineers for several years, the Pentium 4 project underscores the difficulties of managing communications among a massive design team.

"A project like this is a huge undertaking, with many hundreds of people, so keeping them all marching together is an enormous task," said architecture manager Dave Sager. "Not too many companies could do this."

Keeping communications flowing among the various groups involved in the project was particularly important in the early stages, since those decisions had a domino effect on subsequent events.

"Some of the principal concepts that went into the rapid-execution engine came from close cooperation between the circuit guys and the architecture team," said design manager Jim Brayton, who has been at Intel since graduating from Yale in 1987. "There was very active repartee; one group said, 'If you do that, I can do this.' In most projects, there is not this close a link between the architecture and circuit groups. They tend to be somewhat removed at this stage."

That rapid-execution engine is a key aspect of the pipelined processor. It consists of arithmetic logic units that run at twice the core frequency, executing certain instructions in half a clock tick to reduce latency.

Not all of the communications that ensued earned high marks from the chip designers.

"There are managers, who have their point of view, and doers, whose point of view looks a little different, Sager said. "As we approached the deadline, everyone was saying, 'We've got to get this done and that done.' Managers 'helped' by making us do reports twice a week and, as things got tighter, by going to daily reports, then twice a day."

Sager, a former Digital Equipment designer who spent some time with ill-fated supercomputer com pany Kendall Square Research, has been around long enough to feel confident taking pot shots at management's proclivity for paperwork. He got involved with the Pentium 4 project in February 1995 along with a handful of others. "We started with a blank white board setting lofty goals everyone [thought were] impossible. It was a small number of people the first year."

Another early participant was Steve Rich, who was designing Power PC chips in Vermont when he saw an article on the Intel project then code-named Willamette. Excited that it would be an entirely new architecture that would double the pipeline to 20 stages to facilitate the "visual Internet" era, he faxed his resume on a Friday evening. On Monday morning, he got a call from Intel.

Rich is a relative newcomer but has been at Intel long enough to have observed a sea change in the way information is shared.

"We made a lot of use of our intranet systems internally," said Rich, a design automation manager. "The net was just budding at the outset of this program. As we adapted it, we put more and more design data up there, with documentation and cheat sheets. You could find all kinds of data.

"The visibility you could get into the project daily was great. We didn't have to wait for status on the runs; we could just find it on the Web right away. All of Intel is on the net now. You rarely see a paper form anymore."

After the early participants set the overall plans for the chip — which initially boasts a 400-MHz system bus — many of the Pentium Pro team joined in, changing the level of work dramatically. One team member estimated that 90 percent of the overall team came from the Pentium Pro project, with the rest coming from outside the company. Most of the engineers stayed for the bulk of the five-year development cycle, making for quite a crowd.

"We started out downstairs in one building, then we moved upstairs," Sager said. "We outgrew that and were set to move into a building that was being built, we think, especial ly for us. But it wasn't ready in time, so we moved to another office.

"We all got pretty good at packing up our offices: One of the challenges was packing a number of boxes in one day. It was a lot of work. After the last move, we were forbidden to come to work for a few days."

Despite the close quarters, team members say they benefited from being kept together. "No one was more than one floor away," noted validation manager Blair Milburn, a Duke grad who joined Intel in 1988. "That way, you could run into someone and have a conversation in the hallway."

At the moment, with the chip's public debut just around the corner, most members of the design team are working in dual roles: They are doing some final work and making improvements on the Pentium 4 design while getting started on new projects. Many hope that they can stay together after the book closes on the Pentium 4 project.

"We have a very strong sense of teamwork at this point," Brayton said. "People like working together. In the cor e of the design team, all of us were on the same floor working together."

That sense of teamwork was bolstered by several moves designed to keep morale high during the long, arduous task of designing a 0.18-micron chip with 42 million transistors.

"About three years ago, the validation team started going to the beach [every] September for a sand castle contest. We're getting pretty sophisticated castles now," Milburn said. "We also got involved in a local fun run, a 7k run, which got everyone outside."

While individual design teams took the initiative for those outings, the company itself sponsored some surprises. "When the first tapeout happened in December 1999, a couple of floors were filled with 10,000 balloons, and there was all kinds of confetti," Rich recalled. "It was pretty impressive. But the janitorial staff wasn't too happy."

One group that had to be kept happy was the executive branch. "The presentations to the executives increased in frequency the closer we got to tapeout," Bra yton said. "As we got to six to 12 months before tapeout, we got a lot of executive attention.

"But at some point, they got the feeling that they had made all the impact they could. As we got closer to tapeout, things moved to more informal conversations."