Moore’s law grew the IT business for 51 years. But today’s anniversary might be its last
Fifty-one years ago to the day, the director of Fairchild Semiconductors’ R&D laboratories wrote an interesting paper.
He described the evolution of computer chips since the 1950s.
In a short text – mere three and a half pages long – he described his findings. He noticed that the transistor count of reasonably priced circuits is growing at a quick pace. The chart of transistor count growth for the last few years was practically a steady upwards line. The complexity of similarly priced circuits almost doubled every year.
As a proper scientist, the director thought ahead. What could this mean for the future? “Over the longer term, the rate of increase is a bit more uncertain, although there is no reason to believe it will not remain nearly constant for at least 10 years,” the thirty-six years old researcher wrote.
His name was Gordon Earle Moore and he just gave the world his “law” that dictated the pace of the whole IT industry for over half a century.
In his original study from 1965, Moore wrote about the transistor count doubling in recent years and predicted that this tempo might last for a while. Back then he probably had no idea that in about three years, he will leave his post of director of R&D at Fairchild Semiconductor. And that he will establish, together with Robert Noyce, the CPU-producing company Intel.
Ten years later, Moore revisited his original research and reformulated his hypothesis to be a bit more sober. This time, his prediction was that further development of circuit complexity will go at a bit slower pace. The doubling of transistor count should happen “only” every two years.
This is how it all started: With a chart of component count per integrated function (in log2)
The chart that started it all looked something like this. In his 1965 study, Gordon E. Moore described the steady growth of complexity in chips in the last years. You can see the original chart on the third page of his text.
This is the final form of the so-called Moore’s law that most people today know – every two years the transistor count (or, in some simplified retellings, the “computing power”) should double. Moore’s very own Intel helped this idea along considerably – for years, the company followed its founder’s prediction as an actual law and introduced new manufacturing processes and other improvements every two years.
The rise and fall of one prediction
Lately, however, it seems that Moore’s decades-old prediction is not cutting it anymore in the current IT industry. The manufacturing processes are hitting the limits of laws of science. Not even Intel is able to introduce financially viable improvement to the process every two years. Intel has recently confirmed this in a document for the US financial agency SEC. Intel’s document stated that the company will stay at the current 14 nm manufacturing node and also at the upcoming 10 nm node for longer than two years.
This year’s anniversary of Moore’s law might very well be its last – as the “law” itself might not be relevant in the upcoming years. As such, there has never been a better time to read the articles that started this whole thing – you can start with the original paper and then continue on to the revised approach Moore penned ten years later.
It might seem that it’s done for. But Moore’s law might still have a tiny chance to survive. Intel’s management stated that they would like to return to the two-year cadence of new manufacturing processes – but they’re still waiting for technology that would allow it. Currently, the most promising tech for further reducing transistor size (and thus allowing for more transistors in the same die-size, improving the transistor count) is the so-called extreme ultraviolet lithography (usually just called EUV for short). However, this technology won’t be used earlier than at the 7 nm manufacturing node, which might still take three or four years to arrive.
While the Moore’s law is not definitively dead yet, it’s been indefinitely put on ice.