Where Moore’s Law Dead-Ends
“What computing rule of thumb,” reads a trivia question on How-To Geek, “predicts the doubling of computing power every two years?” It’s a tougher question than you may think.
This article originally appeared on CMSWire.com, a publication of Simpler Media Group, and is no longer accessible from that site.
Fifty years ago this weekend [counting back from April 17, 2015], an electronics engineer named Gordon E. Moore who had co-founded Fairchild Semiconductor, and who had moved on to lead a kind of startup firm with the strange sounding name Intel, published an article in Electronics magazine (PDF). In it, Moore shared an observation that the costs of producing more highly integrated electronics was declining at a predictable rate.
The article, with the beautiful title, “Cramming More Components onto Integrated Circuits” (perhaps Moore was also an SEO visionary back then) suggested that integrated circuit manufacturers may actually be compelled to find new and innovative ways to combine the various parts of an integrated circuit into ever smaller spaces over time, in order to take full advantage of those declining costs.
“Integrated electronics will make electronic techniques more generally available throughout all of society,” he wrote, “performing many functions that presently are done inadequately by other techniques or not done at all. The principal advantages will be lower costs and greatly simplified design — payoffs from a ready supply of low-cost functional packages.”
For the 10 years that followed, various corollaries would be attached to this observation, by Moore and others. Transistor count became one way of measuring, on a linear scale, the rate at which ICs were being minimized. A check of the most commonly produced processors over the last 44 years does indicate that transistor count has essentially doubled over that time.
In an interview 10 years ago (PDF), Moore said that basic 2x/2 years calculation had been “blindly extrapolated.” Moore has always been modest about his own work, but that’s also because he has always been accurate.
It is indeed the most brilliantly revolutionary observation made about manufacturing processes in the 20th century. It enabled the entire electronics industry to adopt a measured pace of innovation, to set reasonable goals for itself, and to meet those goals. It gave Intel a way to produce roadmaps for itself, looking maybe five years ahead, that were predictable, achievable, and unsurprising.
But that’s not what the trivia question asked, is it? Specifically, the phrase it invoked was “computing power.”
What Power, Where?
This is where the whole Moore’s Law thing ventures deeply off course. We want so badly to have a unifying principle of business — something that enables us to adopt a measured pace of innovation so we can maintain low costs and manage reasonable margins. We admire how Gordon Moore accomplished this for microelectronics.
But “computing power” is the fuzziest concept in all of marketing — more so than “the cloud,” much more so than “happiness.”
Power, the way it is generally perceived, is not a scientific observation but a subjective supposition. The growth of power is a principal theme of marketing. The suggestion that power grows on its own, or that it multiplies as a force of nature like reality shows on cable, was never made by Gordon Moore.
For the last five decades, Intel has happily chased the “Law” inspired by its founder. It has made no effort, especially recently, to camouflage the fact that the pace of innovation established by Moore’s Law requires effort.
Ten years ago, Intel and its then-competitor AMD had come to the realization that it was no longer possible to achieve this doubling of power every two years the way they had been doing, or at least the way they had characterized it in the press: by advancing clock frequencies.
As I wrote for Tom’s Hardware back then, they were forced to implement the concept of parallelism — breaking down tasks into components, and processing them simultaneously — in order to resume presenting themselves as the doublers of power.
Gordon Moore had observed in 1965 that the condensing of microprocessors would produce heat, and noted that methods would need to be conceived for overcoming that barrier. So as I wrote for Betanews eight years ago, Intel devised a new methodology (IBM followed suit the very same week) for etching transistors into ICs in three dimensions.
When Intel explains these innovations to me, they always, without fail, preface their stories by reminding me of the need to chase Moore’s Law. Without continued innovation, the company says, they actually cannot double the power of things every two years.
But is Intel chasing some measurable goal? Each time the scale by which power is measured runs up against its physical limits — the components are too dense, the heat is too great, the power consumption is too inefficient — Intel innovates its processes, and then changes the scale by which “power” is judged.
This is not a wrong thing to do. In fact, this is exactly what Gordon Moore suggested in the original 1965 article: Automate the innovation process. Specifically: “Perhaps newly devised design automation procedures could translate from logic diagram to technological realization without any special engineering.”
“If Gordon Moore’s suggestion was about anything, it was the opposite of disruption. It was about peering into the future, seeing the patterns of change, observing them and deriving from them constants that we can use to pace ourselves.”
The Perception of Power
The error has never been in Gordon Moore’s original suggestion. Where Moore’s Law dead-ends, time and time again, derives from the realization that the “Law” is but a corollary of that suggestion, an observation about how it could be applied in 1975 — one made, by Moore’s own admission, by a Cal Tech professor named Dr. Carver Mead. It was Mead who came up with the whole “Moore’s Law” thing.
And from there, the “Law” has taken on a life of its own. It has been used to measure the evolution of modern culture. Salesforce CEO Marc Benioff leverages Moore’s Law at every opportunity, usually in explaining why it is that things must double. Benioff calls the “Law” a disruptive influence in marketing.
No. If Gordon Moore’s suggestion was about anything, it was the opposite of disruption. It was about peering into the future, seeing the patterns of change, observing them, and deriving from them constants that we can use to pace ourselves.
It is not the idea that power must increase. For let’s face facts: The ratio of the number of transactions computers can perform, to the number of accomplishments human beings can achieve with them, grows at a rate which dwarfs the expansion of the universe. We are nowhere near the point in our history where the true power of our technology doubles every two years, or even within our lifetimes. If anything, we have built more inefficiencies into systems, and have made greater excuses for them.
The periodic redefinition of “power” as it applies to products and services is, as many marketers will attest, a key function of marketing. Throughout the 2000s, Moore’s Law was repeatedly confirmed by estimates of the growing power of desktop computers. That power isn’t growing any more, and frankly wasn’t really growing through much of the 2000s, at least insofar as the actual work product of computers was concerned.
But that didn’t matter, because we had stopped measuring. We use iPads now, and look how powerful those are! We store our media in the cloud, and its power is conveniently immeasurable.
(For more than one article, I’ve joked that when a product such as a computer or a laundry detergent or a congressperson stops increasing in power as measured on its current scale, the number of alternative methods conjured for consumers to estimate power differently, doubles every two years.)
Gordon Moore’s suggestion was never about the subject of that trivia question. Nor was it about disruption, motivation, or the customer experience. It was a realization that production costs are reduced in an industry when customers expect the yield of those products to increase, and that there was a simple proportion between the two.
That same proportion may not apply to other industries. It may change over time. As it changes, our current methodologies may dead-end, and we may be forced to adapt and start over. But whoever we are and wherever we work, we can capitalize upon progress when we can keep it under control.
If we can just stop trying to rephrase Moore’s words into something suitable for a greeting card or a bumper sticker, we might just learn something from them.
