The characterization of a microprocessor design in a given process will tell the product engineers the range of maximum clock speeds that they can expect from parts coming out of their fabs. An important factor to the speed setting process is the specified acceptable range of supply voltage and temperatures supported by the device. The maximum supply voltage is defined by studies of accelerated aging of parts made in that process. In general, the higher the supply voltage, the shorter the average lifetime of parts will be for a given temperature. This is due to higher electrical stresses within the transistors and larger currents flowing within the interconnect (current consumption generally increases proportional to supply voltage). The accelerated aging studies yield data that is extrapolated to voltage and temperature ranges of normal operation.
For example, the engineers might calculate that a maximum supply voltage of 2.20 Volts combined with a maximum junction temperature of 105 degrees Celsius will give the desired lifetime of ten years (or whatever) for 99.99% of parts. The minimum and nominal supply voltages are usually determined by scaling the maximum voltage according to industry practices. For instance, it is common to specify supply voltages to a plus or minus 5% or 10% range. Some microprocessors actually instruct their power supply circuit to provide them with a specific voltage using a binary encoded control value on four or five output pins. In our example, a 10% tolerance supply would be 2.00 Volts plus or minus 0.20 Volts. The combination of a 1.80 Volt minimum supply voltage and maximum junction temperature of 105 degrees can then be used to predict the speed distribution of parts that come out of the factory. The parts are tested and separated into a number of speed grades (i.e. binned) for sale at different prices.
The actual determination of speed grades that a microprocessor will be sold in is an exercise both in financial and operational analysis, and in marketing. The goal of the exercise is to set the speed grades and their prices to attract buyers to each speed grade roughly in proportion to which these parts come out of the factory (i.e. align supply and demand), and yield the most revenue and highest margins for the business. But sometimes, competitive pressures are so high that the clock rate of the fastest speed bin is set unrealistically high. When this happens a shortage of the high-speed parts develops while the MPU vendor is awash with slower binned parts.
Why Should It Ever Run Faster than Spec?
As thousands of wafers of microprocessors in a given process are churned out month after month, the process engineers learn how to reduce the variability in effective gate lengths, gate oxide thickness, etc., using statistical process control and other measures. This allows them to push the process center towards faster and faster devices without incurring the risk of building unreliable parts. Also, microprocessor designs are often “tweaked” as time goes on to speed up critical paths. As a result, the yields in the higher speed bins gets better and a shortage of the slower parts develops as supply and demand gets out of synch. Over the long term this is usually addressed by introducing a new high speed grade part and forcing out the slowest part through price compression (reduce the price of the second slowest speed grade to level of the slowest speed grade). But over the short term, microprocessors are often sold as slower speed grades than the devices are actually physically capable of. This is one opportunity for overclockers to beat the system.
Another opportunity exists in the way that computer and microprocessor marketing managers like to spread out performance increases smoothly over time. New chip designs and semiconductor processes lead to huge step increases in performance, followed by years of relative slow performance increase from process and design tweaks. The MBA types like to artificially slow down these technology driven step increases in performance so as to spread the increase over the two or three years before the introduction of the next new design or process. As well as allowing the maximum profits to be realized from buyers, this gradual approach is also necessary to provide the time to cut over from old MPU design and/or process to the new one. Otherwise, the older parts would suddenly be unsellable and production of the new parts couldn’t be ramped up fast enough to meet demand. So there may be an opportunity to overclock even the maximum speed grade of a new processor at its introduction due to the clock rates being artificially held low early in the product cycle.
Finally, the maximum clock rating for a microprocessor is set to allow the device to reliably operate under the worst case conditions specified in the datasheet. If your system can consistently keep the processor significantly cooler than the maximum designed operating temperature and supplied with more than the minimum operating voltage, then there is room for operation at a somewhat higher than rated clock frequency, even if the device barely meets specifications.
Discuss (One comment)