Facing customer skepticism, AMD remakes its chip-design engineering process

2009 InfoWorld CTO 25 Awards: Jeff VerHeul

2009 InfoWorld CTO 25 Awards

Jeff VerHeul
Corporate vice president of central engineering


Sometimes, a CTO's job is not much fun. Rather than focus on bringing innovation, you have to clean up a mess instead. But a great CTO can take the cleanup work and use it to introduce innovation. Case in point: Jeff VerHeul, corporate vice president of engineering at chipmaker AMD.

Although well received as a product when in shipped in 2007, AMD's "Barcelona" quad-core chip was a failure in terms of execution, having suffered from delays that caused customers to question AMD's prowess in the emerging multicore chip market. "Delays in the release of Barcelona caused some OEM partners to question AMD's ability to execute on new, cutting-edge product introductions," VerHeul notes. In the hypercompetitive chip market, it's difficult to recover in time for the next round of chip advancements.

[ Discover how the lessons learned from the 2009 InfoWorld CTO 25 Award winners can help your IT efforts. ]

Fixing the engineering management woes that let Barcelona's development slip so the next major chip effort -- "Shanghai," the planned Opteron successor -- would not be delayed fell to VerHeul, who also had to drive technology innovation.

To fix the management deficits, VerHeul realized that the multiple engineering fiefdoms within AMD muddied responsibilities and allowed gaps to occur in the product design. So he consolidated the responsibility for testing the new processor under one person, who assembled a cross-functional team across all engineering groups (such as design and validation) that met daily for the next nine months. In some instances, the team realized it could move faster by slowing things down. For example, at one point, it delayed a tape-out (the creation of the circuit image used to produce a sample chip) by a month. The result was a more solid design that saved the debug team from investigating thousands of false alarms, which sped up the overall process.

VerHeul also brought PC makers into the design and testing process much earlier. Before the Shanghai effort, AMD engineers would debug processors internally for months before releasing hardware for OEM testing. VerHeul changed the procedure so that OEMs started testing when AMD did, validating Shanghai on next-gen PCs in lockstep with OEM partners from the beginning. This let OEMs quickly validate their new platforms and shorten their own system launch dates.

While reworking the engineering management process and helping speed OEM design, VerHeul and his team also focused on advancing the chip technology. As a result, when it shipped in late 2008, Shanghai could be used in 2P to 8P servers with a single architecture, while maintaining socket and thermal compatibility with older designs. Plus, Shanghai provided up to 35 percent more performance on average than Barcelona and 35 percent less idle power consumption.


Copyright © 2009 IDG Communications, Inc.