After almost two weeks of wrangling with its legal department, Nvidia has finally answered some of our questions about its GPU and chipset failure problems. First, the company sent us a statement addressing the allegations of AMD's Packaging and Interconnect Director, Neil McLellan. As we wrote earlier this week, McLellan claims AMD has a superior chip package design because it uses a specific set of materials, including eutectic solder bumps. In his view, Nvidia chips have failed because they use (supposedly) more fatigue-prone high-lead bumps and because, he asserted, Nvidia cares less about packaging technologies.
Here's Nvidia's written response:
NVIDIA has a world class chip operations team, and has delivered over 1 Billion devices (and over 1 trillion bumps) over 14 years, in the most advanced processes, to the most demanding customers. NVIDIA is the leader in the graphics industry in innovation and has delivered technology over the years that companies like ATI and Intel have benefited from.
In his recent commentary on chip packaging, Mr. McLellan makes a number of speculative assertions about NVIDIA's people, products and philosophy. In his interview McLellan asserts that High lead bumps are more prone to fatigue. What he fails to note is that AMD currently uses High lead bumps on their CPU line -- a device well known to undergo high thermal stress, and also go through lots of power cycling.
The choice between High Lead and Eutectic is complex. There are trade-offs in using one vs. the other, as even Mr. McLellan points out. The electromigration issues associated with printed eutectic bumps can affect long term reliability of a high current device. Electromigration is when a high current causes metal to separate over time, and creates an open circuit. This is one of the main reasons why so many devices are still manufactured with High lead bumps today.
In fact, 10s of billions of semiconductor devices have been shipped with High Lead bumps by world class companies including AMD, Intel, IBM, Motorola and TI.
While McLellan implies that AMD is unique in its use of a "power redistribution layer," this isn't true. In fact use of a power redistribution layer is industry practice, and has been used in every flip chip GPU NVIDIA has shipped.
NVIDIA is committed to delivering lead free devices by the 2010 requirement. Our engineering team has been working on this important initiative for the past 18 months, and is fully engaged in this effort with our manufacturing partners.
NVIDIA uses industry standard packaging material and we have passed all industry standard (JEDEC) component package qualifications. We stand behind our products and we will continue to work with our partners to ensure the best visual experience.
After looking over the statement, we got on the phone with GeForce General Manager Ujesh Desai and GeForce Senior VP Jeff Fisher to ask some follow-up questions. What exactly causes Nvidia chips to fail in the first place? Can the same failures occur in desktops, and is that what we're seeing in the HP systems we talked about earlier this week? What of the GeForce 9400M "motherboard GPU" in Apple's new notebooks?
Predictably, we couldn't get a more concrete answer on the specifics of the failures. Desai was, however, willing to point out that the failures only affect a "small percentage" of notebooks, and the problems depend on a combination of factors, so "you can't just . . . point back to our chip."
Fisher was notably chattier on the topic of potential desktop failures:
There is no evidence that this issue exists in desktops as we know them. And in fact, Mr. McLellan has no evidence to even imply that. The fact is that lead bumps—he's saying that lead bumps will fail, and therefore you should expect to see failures on everything, and that's completely out of balance from an educated operations guy like he is. . . . I think most industry people would say lead bumps are not a cause of failure and are in fact very reliable. And his soda-can analogy and attempt to drag in desktops is irresponsible from our view and a huge reach.
About the failing HP systems, Desai specified, "It's not 38 different systems, it's actually a single design, and the model numbers that were reported . . . are actually model numbers that refer to different configurations of the same product." He went on to say Nvidia is "working closely with HP to determine if or how the Nvidia chips are even involved in the failures."
So, looking forward, what can we expect from the GeForce 9400M chipset in Apple's new MacBooks? Fisher stated plainly, "You can rest assured that Apple has been aware of all of the science that we've developed around this issue and would not be launching the most important product in their history with a product they felt was at risk." Desai later added that Nvidia is "taking the necessary steps to ensure that all the Nvidia chips currently in production don't exhibit this problem."