Inside Intel: From Silicon to the World

A Menu of Technologies

Keep in mind that the main goal at the Intel Labs is to create a menu of technologies that the product and design engineers can then select from to include in the products that they’re making for the future.  For example, IAL engineers may design a double-pumped ALU in order to test out a new circuit theory, this double-pumped ALU may eventually get included (in one shape or another) in a processor such as the Pentium 4. 

We use the ALU example because that happens to be a very common thing to experiment with in the labs.  The reason being that it’s pointless to spend design time working on circuits that have no purpose, but by creating a useful set of logic such as an ALU not only do the technologies that went into its production get tested, but the ALU itself can then be later used in products. 

One such technology that the engineers have been working with and were eager to discuss upon our visit was the idea of adaptive body biasing transistors.  There is an incredible amount of math and analog circuitry that goes into the idea of body biasing transistors but the simple $0.10 version of the story is that by controlling the voltage applied to the body of a transistor you can cause the transistor to switch either faster or slower.  By applying a positive body bias you can make the transistor switch faster, normal bias will make the transistor switch at the same speed, and zero bias will make the transistor switch slower. 

Adaptive body bias is the dynamic adjustment of body bias on transistors depending on how fast or how slow they’re switching. This seems like a simple thing but its applications are numerous.  Remember that the speed of a complex circuit is dependent on the speed of its slowest part.  For example, if you have a circuit whose transistors are capable of switching at 10GHz but a small portion of them can only switch at 8GHz, your circuit (e.g. CPU) will be limited to 8GHz.  Using by applying an adaptive body bias to the slower transistors it may be possible to speed them up to the speed of the rest of the circuit at 10GHz.  This can have a huge impact on yield as it will be easier to get more transistors to run at higher frequencies.  Especially moving forward as CPUs begin to have hundreds of millions of transistors, the likelihood of some not switching as fast as the rest increases.  Adaptive body biasing will help tremendously in situations like these.