A few simple facts about silicon-germanium transistors
(Some of you may remember me as Dennis Sellers' primary analyst resource for his series of articles entitled "The Myth of Megahertz" at
http://maccentral.macworld.com/news/0104/09.megahertz.shtml )
IBM's recent announcement involves a simple "ring oscillator", a circuit consisting of just a few transistors that generate a clock signal. This original high-frequency signal is then divided down into a stable lower-frequency signal that can be used by other chips.
The base region of the transistors used in this circuit are made from a silicon-germanium (SiGe) alloy with a very low percentage of germanium. The germanium increases electron mobility in the silicon, allowing circuits to operate substantially more quickly.
The SiGe layer in the chip is first made from pure silicon, exactly as would be done for any other chip. Germanium atoms are then implanted into the layer. The electrical properties of this alloy depend on the exact proportion of germanium, and it is very, very difficult for IBM to get this proportion right for very many transistors on a chip.
As I wrote for our newsletter,
Microprocessor Report, in May of 1996:
The new technique is not as useful for large digital components such as microprocessors, because it improves only bipolar transistors, not the FETs used in CMOS devices. SiGe transistors could be used to extend the life of BiCMOS design, which most vendors have turned away from due to the minimal gains for bipolar devices at geometries below 0.35 microns. IBM Microelectronics has no immediate plans to develop SiGe microprocessors, but the new technology could allow faster, less power hungry versions of current bipolar components.
IBM does another round of publicity for its SiGe process every three years or so. Every time, news services and end users leap to the conclusion that faster microprocessors are on the way. I don't know if this is what IBM wants people to think, but I wish they'd be more careful.
There are other silicon vendors, such as Motorola, using similar techniques to build high-speed transistors for communications devices, but they all have the same limitations. None of these methods can be used to create 100GHz microprocessors.
. png
Peter N. Glaskowsky
Senior Editor, Microprocessor Report
Columnist, Electronic Business magazine