Bad Transistor Causes Billion Dollar Mistake
Sandy Bridge chipset architecture
Sandy Bridge with P67 chipset
Sometimes it is a good idea to get all the bugs out of the way before investing too much money in a project. Unfortunately for Intel, a bug did slip into the Sandy Bridge P67 chipset in the form of a bad transistor in the controller subsystem (Fig. 1). Sandy Bridge is Intel's 32nm second generation of Core processors (Fig. 2). Intel was up front about the issue in a recent press release (see Intel Identifies Chipset Design Error, Implementing Solution).
Sandy Bridge was being shown with a lot of fanfare at CES 2011. Cost estimates are putting the total expense for Intel at around a billion dollars. That includes $700 million for handling the recall and $300 million in lost sales.
The problem occured in Intel’s 6-series H67/P67 chipsets that has two sets of SATA ports. One set handles four 3 Gbit/s drives and the other works with a pair of 6 Gbit/s drives. The transistor of concern is located in the PLL (phase lock loop) clock tree of the 3 Gbit/s controller. The circuit was biased at too high a voltage for the design and this resulted in an excessively high leakage current. This in turn changes the system's characteristics and causes the controller to fail. The other controller is unaffected as well as it has its own PLL.
It is possible to remove the voltage from the transistor but this is a major board change. Unfortunately a software fix will not work. This is why distributors have stopped shipping motherboards based on these chips.
The problem did not occur in the initial A-stepping chips that were used for design and test. The issue is with the B-stepping chips that were shipped in bulk and found on commercially available motherboards. The updated chip will likely have a C-step or D-step designation.
The problem does not always show up and tends to occur when the system is stressed. Intel found the problem while doing these kinds of tests although not soon enough to prevent many motherboards from being constructed with the bad chip.
Issues and Opportunities
The affected chipset is not the only one that works with the Sandy Bridge processors. Likewise, there are many applications where the problem is a non-issue. For example, many embedded or mobile systems will be supporting only one or two drives. These are likely to be connected to the faster 6 Gbit/s ports that are unaffected. This class of system will run just fine with the existing chips.
The bigger issue for developers will be dealing with the lack of systems based on this chipset. Sandy Bridge processors will still be available and other chipsets can be used. For board designers, this means waiting for the updated chip or redesigning a board to use another chipset.
System designers have more flexibility since there will be a number of alternatives available although not necessarily with price/performance characteristics of the problem systems. This may open doors for solutions based on other chips like those from AMD and VIA Technologies as well as chips using Arm's architecture. The amount of delay before a particular company gets their hands on the updated chips or boards using these chips will be the main issue. Unfortunately this amount of time may be significant with many variables including when a fix becomes available and where the company is in the line to get new chips or systems.
Of interest will be AMD's new dual core G-Series processors (see APU Targets Embedded Applications). This uses AMD's APU (accelerated processing unit) with a built-in Radeon graphics controller SIMD computational support in addition to handling display chores.
There is also VIA Technologies' dual core VIA Nano X2 (watch VIA Nano X2 Processor and VL810 USB 3.0 Hub Controller).
The AMD and VIA Technologies processors are paired with their own hub controllers.
So how has this news affected your plans?
Editors note: Check out Bad Transistor May Not Cost A Billion Dollars for Intel's response.
