Overclocking – just say no

You don't have to look far on the Web or in the more geeky IT mags to see loads of "mine's faster than yours" stories where awfully clever people are overclocking systems to within an inch of their lives. Celeron 300As are one of the overclockers' favourites and it isn't unusual to see boasts of these poor little processors being run at more than double their rated speeds. Chip manufacturers in general and Intel in particular have long taken the view that processors may run at speeds higher than their ratings, but then again, they may not -- and as soon as you've tried squeezing more performance out of your CPU by nefarious means, you're on your own -- no warranty. While this probably isn't a major setback for a chip costing less than eighty bucks, blowing up a Pentium III or even an Athlon 600 is going to put the hurts on your bank balance, big time. In fact an Intel insider once memorably told The Register: "Hell, run 'em as fast as you like. When it blows up we'll be happy to sell you a new one." Checking out the favourite overclocking sites like BX Boards (which we rate, incidentally) will introduce you to the wonders of super-duper cooling systems costing more than the CPU they're designed to protect while you attempt to run it at 2GHz or whatever. But they won't prevent the CPU failing -- you cannae change the laws of physics, Jim, as a famous engineer once said. There some nasty problems you're risking once you start upping voltages inside any integrated circuit. While technology sat at 0.5 micron and 0.35 micron, these problems didn't often manifest themselves due to the sheer butchness of the chips -– traces were pretty huge in semiconductor terms. But now that even the naffest CPUs are built at 0.25 micron and the state of the art is a skinny 0.18 micron, we're in what's termed Deep Submicron (DSM)territory. Down here in Tiny Town, scary things start to happen. A lot of the potential problems can be designed out before the silicon hits the streets, but the headroom is reduced and devices become far less tolerant of running at higher voltages than they were originally designed for. Chip designers call this problem design integrity -– reliability to you and me -– and according to those nice chip design people at Cadence, the three main problems are (pay attention at the back, this is complicated): Electromigration This is the most likely reason an overclocker will kill a processor. When the DC current in a line is too high, the metal grains that make up the wire are physically pushed aside by the electron wind. The longer you run the chip at higher than design voltages, the more the metal is distorted. Eventually it gives up the ghost and the circuit fails permanently. Hot Electrons Again caused by overvoltage, when there is a high voltage between the source and the drain of a device, a high electric field is created and electrons accelerate, damaging the oxide and interface near the drain, changing the transistor threshold and mobility. In an N-transistor, the gate is always positive and the shift is always in the same direction. Eventually the threshold moves to a point where the transistor no longer switches and is effectively dead. This problem is exacerbated by the move to smaller technologies as, although device voltages are reducing as sizes come down, they aren't reducing in proportion to the device shrinkage, leading to higher field strengths compared with older 0.35 and up devices. Wire self heat Known to its friends as signal line electromigration, this is caused by frequently varying thermal conditions. The wire heats above the oxide temperature as pulses go through it, due to the power dissipated by the wire itself. This causes mechanical stress and eventually, the wire fails. So now when your Celeron 750 or K6 900 stops working, you know why and you'll only have yourself (and maybe those awfully clever overclocking Web sites) to blame. ® OK, overclockers. Pete Sherriff has laid down the gauntlet. You can reach him here