« Fake uterus | Gene Expression Front Page | Iceman died fighting & palaeo-fiction »
August 13, 2003

Diamonds are a CPU's best friend?

This article from the print edition of Wired about synthetic diamonds is pretty cool. Synthetic diamonds seem to have two main benefits:

  • Long term: Replace silicon chips and get around some of its inherent limitations (silicon obviously can't handle as much heat as diamonds).
  • Short term: Kick the balls of the "blood diamonds" industry that gave us Charles Taylor et al.

Posted by razib at 11:55 AM

I'd take those predictions of diamond CPU's with more than one grain of NaCl. Neither this nor any other article I've seen discusses the really important numbers that determine how a semiconductor will perform in electronics, but I think I can get a ballpark estimate from the periodic table position relative to germanium and silicon. That is, column IV is:


In the form used for semiconductors, Si and Ge both have the same crystal structure as diamond (all the atoms linked by tetrahedral covalent bonds), but they are progressively easier to grow into crystals and softer as you go down column IV. (Sn = tin is a metal, where valence electrons detach from the atoms and flow freely through the material, so the diamond structure doesn't occur.)

Germanium was the first material used for transistors. It has a very low bandgap energy, which means a low voltage drop when current flows across a PN junction in the forward direction (that is, positive is connected to the "P"-biased region). This is good, but Ge also leaks quite a lot of current when the diode or transistor is biased so it should be off. It's pretty hard to get consistent circuit operation when the switching elements won't actually switch off.

So they moved to Si, with a higher forward voltage drop (0.7V), but negligible reverse leakage current. Si transistors also are slower than Ge transistors of the same size and construction, but this problem has been overcome because Si's other characteristics allow shrinking the transistor size, varying the transistor design, and combining lots of transistors on one chip of Si. (Ge parts are still used in some extremely high-speed but simple circuits for communications.)

One bad side to the higher forward voltage drop is that every electron flowing through a PN junction gives up 0.7 electron-volts of energy, normally as heat. Or 1 micro-amp of current flow through one junction = 1 micro-watt of power turning into heat. Multiply that by 35 million transistors and you've got as much heat generation as a small soldering iron. Pentium chips with far more transistors than this survive even with massive heatsinks and fans only because only a few percent of the transistors are used in any one cycle, and these only for short bursts of current rather than conducting steadily.

I expect C (diamond) would have a substantially higher bandgap energy, for several volts forward voltage drop. That is, a diamond Pentium would have several times as much heat to dispose of. Maybe they'll find ways to work around this in 20 or 30 years, but I doubt anyone's going to be putting in the research dollars. Si works so well already. Also, it's much easier to grow Si crystals and to slice them into wafers - I think they use diamond saws for this, what would you use to slice diamonds?

There definitely are potential uses for diamond in electronics, just not for computers and other high-density logic circuits.

1) Diamond has extremely high heat-conductivity. (I guess that the very strong covalent bonds connecting atoms means that when heat makes one atom jiggle, the motion passes on to the other atoms very quickly.) So a diamond film could make a good interface between the Si computer chip and the heatsink. For this application, you wouldn't dope the material to create holes or spare electrons, but just keep it as pure and high electrical resistance as possible. So the diamond could be both a heat-spreader (to increase the size of the contact area to the heat sink) and an insulator. AFAIK, the next-best solid materials for heat-spreaders are silver, gold, copper, and aluminum, but these often require a layer of insulator so they don't short out the electronics, and the insulator is extra resistance to heat flow.

2) A good many applications require voltages that are a challenge to Si parts. For a medical diagnostic system, my employer builds 1,000 VDC photo-multiplier tube power supplies. The rectifier diodes for this are almost as big and expensive as the transformer; I suspect that they are actually whole stacks of Si diodes. I'd expect diamond to naturally handle high voltage much better. Diamond transistors could help with other high power and voltage applications, maybe things we now still use electromechanical switching for like HV power transmission lines.

3) In LED's and laser diodes, part of the energy lost when electrons drop across the bandgap voltage at a PN junction is converted into photons. Each photon must be formed from the energy lost by just one electron, so an LED built from Si can only output infrared, at less than 0.7ev per photon. For visible red light, you need something like Gallium Arsenide tailored to a bandgap around 1.5V. Green is a little, higher, blue is so high it's only been achievable for a few years. The materials used in visible LED's are exotic, expensive, and fragile, and it's quite difficult to get enough power density for laser action without the diode self-destructing. Diamond ought to give you blue or even ultraviolet light, from a material that may be expensive but is stronger than steel and should be capable of extremely high power density. Think blue or UV lasers for higher density DVD's. Or you could coat the diamond LED with a phosphor (like flourescent lights) and get white or any other color out - a lightbulb that's over 50% efficient and never burns out.

Posted by: markm at August 16, 2003 06:24 PM