The Race For A New Game Machine:. David Shippy

require significant monetary investment—top-notch salaries, employee benefits, office space, silicon test chips, lab space and equipment—in order to mature into hardware. Sharing that finite pot of gold with its sister division simply reduced either’s ability to pursue homegrown innovation. Parochialism was indeed entrenched in the IBM culture and presented a never-ending source of contention. As with the mainframe processor wars of the 1980s, the group awarded the PlayStation 3 processor would secure funding from corporate IBM to staff up a large engineering team. So it all came down to job security again.

      An IBM Fellow (the highest rank on IBM’s technical career ladder), Rick Baum, drove a task force to resolve the processor roadmap for the entire IBM corporation. The words “task force” made me cringe. This method of problem solving was a relic from the old IBM culture, and I felt it was a huge waste of time and money. It’s a committee of high-level engineering experts, pulled from various disciplines within IBM for the express purpose of solving some problem du jour. They would meet on a regular basis, with each member reporting the results of their various homework assignments. Grouped into opposing camps, each side would attempt to persuade the other that their engineering position was the right one, and it was next to impossible to reach a consensus. As I met with Baum’s task force, I wondered once again if jumping back into a big corporate environment had been the right decision.

      The conclusion, if there was to be one, could be very important to the STI Design Center, so I tried to stay focused. Some members of the task force lobbied for one common processor core to cover both the high-end server space as well as the high-end game space. Kahle was pushing hard for a separate game core.

      Kahle presented his case to Baum and the task force: “The server requires something entirely different than the game console does. I need a small, simple core that focuses on high frequency, without all of the baggage that the server core will require.”

      The Server folks presented their own arguments, though they didn’t appear to have any good technical data. They just wanted to own the core for bragging rights, and they wanted funds to sustain future engineering jobs. Kahle told me later, “Baum doesn’t really like me much, but I’ve got money coming in from Sony and Toshiba. I’ve also got some important executives backing me because I’m going to help pay for our new fabrication facilities by filling them with high-volume chips.”

      The task force fizzled out, as I predicted, and Kahle got the green light to develop a new PowerPC game core. I was glad the task force didn’t last too long because I desperately needed to fill some of the key leadership positions on my team.

      I called some of my old buddies from my Power4 experience. Most folks were already committed to other critical IBM server chips. However, I lucked onto one veteran engineer, David Ray, who was willing to talk to me. David and I worked together on several previous IBM microprocessor designs. He owned a small piece of property in the hill country outside of Austin, so I teasingly called him “Farmer Dave.” Everyone knew he would rather be building a barn than building computer chips. He was a quiet, crusty engineer who never seemed particularly happy and who especially hated managers; however, he was sharp and always delivered high-quality designs ahead of schedule. He earned the highest respect from his peers and from the people he led. Thankfully, David was in-between projects right then and was fairly unhappy with the Server Group management, so I invited him to meet me for lunch in the cafeteria.

      As soon as we went through the food line and found a table, I jumped right in. I told him the truth: “Dave, I’ve got the deal of a lifetime for you. How’d you like to work on a start-from-scratch design for a supercomputer-on-a-chip for a game machine? It’s going to be extremely difficult work, we’re severely understaffed, so we’ll be running lean and mean for a while, and it’s all sort of top secret. How’s that sound?” I laughed, half expecting him to respond with Are you crazy?

      Instead, his eyes lit up. He leaned forward, rested his elbows on the formica-topped table, and said, “When do I start?”

      He looked serious, but my proposal sounded so ludicrous to my own ears, I wasn’t sure if he was just pulling my leg. “Well, you’ll be my first technical leader, so the sooner the better,” I said.

      Turned out Farmer Dave was serious. The challenge hooked him as much as the start-from-scratch opportunity. Before lunch was over, he accepted the position, and we made plans for his transition from the Server Group.

      David Ray also pointed me to my second technical lead, Jim Van Norstrand. David and Jim worked together on the Power3 microprocessor. Jim was another twenty-year veteran with a wealth of design and management experience under his belt. I knew I was lucky to have him on the team. He was an expert logic designer who was intimately familiar with the instruction unit, the most complex and critical subcomponent of a microprocessor. An instruction unit is responsible for handling low-level software instructions running on the processor. It fetches the instructions from memory, decodes them, and issues them to functional units, which then execute the operations specified by the instructions. The instruction unit is like the head cook in the kitchen who fetches and reads the recipe, while the kitchen staff (the functional units) executes the instructions handed down by the cook. An inefficient instruction unit could break the back of an otherwise reasonably designed processor and limit the ultimate speed it could achieve. Van Norstrand became my second in command, assuming a significant share of the technical responsibility for the entire microprocessor core.

      Jim Kahle recruited Tony Truong (pronounced Trong) back into the IBM fold before I got there, so I inherited Tony as my third technical lead. We had worked together on a previous project in the Server Group, but he left IBM in the late ’90s about the same time I did. I was glad to see he had also returned, and I knew I was lucky to have him lead the design of the PowerPC’s memory subsystem. Tony’s strengths lay in his work ethic and his deep knowledge of memory subsystems. He could hone in on a troublesome area of the design that was difficult to verify, and come up with a way to test it. It might take him several grueling twenty-hour days in a row, but he would finish it, and we would end up with a stronger design because of his work. One of the things I loved about working at IBM was the cultural mixture. Tony was from Vietnam, and I could always count on him to find the best Vietnamese food in Austin.

      With these strong team leaders in place to help me, I turned my focus to the organizational configuration. I created a two-in-the-box structure to lead the team. With this two-leader configuration, an IBM second-line manager and I ran the large PowerPC team together. I focused on the day-to-day technical work as well as some of the project management aspects of the project. The second-line manager focused on personnel and administrative issues as well as the remaining project management issues. It was a divide-and-conquer approach.

      Previous project leaders at IBM had organized their teams around interdependent functional disciplines—for example, logic design, verification, or circuit design. Experience already taught me that this type of organization created too much of a “throw it over the wall” mentality when problems created by one team were passed on to another to solve. I wanted self-contained entities responsible for all aspects of the design from start to finish. I carved up the microprocessor core into functional units and created teams around those units. Each team was responsible for delivering a final physical piece of silicon (hardware), which was logically and physically verified and complete. Each member of that team worked toward one common goal—functional silicon. It was not acceptable for anyone to say, “Well, I did my piece”—meaning they washed their hands of any problems—because they were all responsible for the end goal. The structure created accountability for everyone on the team. I also created a two-in-the-box leadership configuration for these teams. In most cases, one leader focused on logic design and verification, while the other focused on circuit design and physical design.

      I was well on my way to creating a high-performance team.

      While I was elated that all three STI partners vowed to provide their very best architects and designers to meet the challenging project goals, I knew it was a promise they would find difficult to keep. Sony and Toshiba faced the larger problem of convincing their top designers to take a two-to three-year tour of duty in the United States. IBM’s problem had more to do with balancing projects and earnings. While they could draw from a vast