The end is near. Or is it? According to some, the beneficial projections of Gordon Moore about the increasing number of transistors per chip over time are slowly coming to an end. But others say, wait, not so fast. Even Moore, a co-founder of Intel, says it may not be over yet for a decade or so.
All this discussion and conjecture comes about as a result of the 50-year anniversary of Moore’s original projection in the April 1965 issue of Electronics magazine. Moore’s original prediction was that the number of transistors per chip would double each year. That was correct, but the “law” was subsequently modified in 1975 and beyond. With some types of transistors and circuits, the number of transistors doubled every two years and every 18 months for others. This has remained consistently true for decades. Today, some chip companies can make over 10 billion transistors per chip. And that trend continues, for now.
As semiconductor manufacturing technology progressed, the processes allowed engineers to make the transistors ever smaller and smaller. With each reduction in size came an increase in speed and lower power consumption. And one could put more transistors in the same silicon area. That, in turn, lowered the cost per transistor. Today an integrated MOSFET costs only micro cents. Huge gains were made with the switchover from bipolar circuits with resistors to MOSFETs. As a result, we got larger and larger memories and bigger and faster processors. That made today’s PCs and smartphones possible with year-after-year performance and feature increases at no cost increase. What a deal!
Now there are signs that the Moore’s law is coming to an end. Transistor sizes are pushing the limits of physics. Today, Intel is making 14 nanometer (nm) transistors whose gate size is roughly 30 silicon atoms wide. Furthermore, the smaller transistors and their closer spacing have created electron leakage that limits performance and increases power consumption. Lowering supply voltages helped for a while, but the limits on that have also been reached. Clock speeds have flattened and the emphasis now is on power consumption reduction.
One of the key limitations of continuing on the path of Moore’s law is the manufacturing process of photolithography. This is the transfer of the design layout from masks to the wafers. This problem has been solved temporarily by the use of ultraviolet light and some tricky double patterning processes. Even that has its limits. Currently, it is estimated that the limit to transistor gate size is about 5 nm. That dead end, if it really occurs, may be as soon as 2020 or so. Who can really say?
Processor manufacturers have overcome the performance limitation by increasing the number of CPU cores per chip. This has allowed performance to increase without further clock-speed increases. However, programming multiple cores has become an issue. As for memory, transistor density has flattened as DRAM and NAND flash manufacturers hold the line for fear of hurting storage time and reliability. USB drives may not get any larger or cheaper.
This is not the end of electronics as we have known it. We will still have big, fast ICs and great products. Yet, we may not see the continuous performance improvements and cost reductions of the past. Moore’s law may slow down to a doubling every 36 months and that trend may continue for another decade or so until the engineers figure out what to do next and how to do it. In the meantime, continue to enjoy the benefits that this trend has permitted.
Hats off to the now retired Gordon Moore for giving us years of a reliable system to predict performance and pricing of electronic products.