Moore’s Curse: not everything is exponential
Moore’s Law (that transistor densities double every 18-24 months at minimal cost) has transformed our world, and continues to do so. But an unintended consequence, according to Vaclav Smil writing for IEEE Spectrum magazine, is that people have unrealistic expectations for technical progress of all kinds. To pick examples from transportation, energy and health care: buses, batteries and biology are only getting better non-exponentially, and that frustrates billions.
“Energy, material, and transportation fundamentals that enable the functioning of modern civilization and that circumscribe its scope of action are improving steadily but slowly. Gains in performance range mostly from 1.5 to 3 percent a year, as do the declines in cost. Outside the microchip-dominated world, innovation simply does not obey Moore’s Law, proceeding at rates that are lower by an order of magnitude.”
Dr. Smil (a Czech-Canadian and Distinguished Professor Emeritus at the University of Manitoba) doesn’t address the issue of why Moore’s Law is so misleading, so I thought I would offer a few suggestions of my own.
First, almost everyone alive in 2015 has spent more than half their life in a world where Moore’s law applied: the first integrated circuit was 1965! After 50 years, we have all been immersed in a world of high and sustained growth in transistor densities, and it has begun to seem normal.
But it isn’t. Most things don’t double every couple of years…and most technologies that do achieve that kind of growth manage it for only 3-4 doublings. Even Gordon Moore only thought it would last into the 1980s at first. For that doubling to have occurred 25 times (and counting!) is almost unique [see below for comments on DNA sequencing and Hard Disk Drive technologies] in any field of technological improvement. And that uniqueness has likely exacerbated our unrealistic expectations.
Next, doubling transistor density has been weirdly useful. It doesn’t just let us make faster chips for computers; it allowed us to make better memories and storage, better screens or displays, better lasers, and better lighting. In the 1950s, each one of those technological functions would have utilised a different core technology. Memories were on tapes, displays were cathode ray tubes, lasers were crystals, and lighting came from filaments or glowing gases. It is nearly magical that semiconductor chips are capable of performing all those functions, and doing so exponentially.
Finally, the ongoing doubling of transistor density isn’t just about faster or more powerful chips. Increased density also allows cheaper ($50 smartphones, anyone?), smaller and more power efficient devices.
In other words, Moore’s Law has driven exponential improvements across multiple axes: over time, across functions, and across many meanings of the word “better.” No wonder that 21st century citizens expect miracles! But we shouldn’t, and we need to ratchet our expectations down. 1-3% annual improvement is likely the rule for almost all technologies. That doesn’t mean that new technologies (such as 3D printing) cannot be incredibly important and transformative…but we still need to not assume they will progress the same way transistor technology did.
(Hat tip to my friend Brian Piccioni, who drew my attention to this article in his weekly blog reading list.)
[Edited to add] My friend, former colleague, sometime Paris drinking buddy, and professional nitpicker John La Bouff commented that I am misusing the word ‘exponential.’ He is correct: ANY technology that improves continuously over time is exponential, whether that annual growth rate is 100%, 42% or 1%.
The problem is that many people use the word exponential to mean “doubles in a short period of time.” If chip density improves by 42% annually, then it will double in 2 years. If batteries improve capacity at 1% annually, they will double in 70 years. And that latter example is NOT what most people think of when they use or hear the term exponential, whereas the former is.
John suggested the word “geometric” instead of exponential. That occurred to me too, but the headline “not everything is geometric” won’t be understood by most readers!
[Further edited to add] Brian Piccioni and Paul Barter at MaRS noted that there are other technologies that have been growing at Moore’s Law kind of growth rates. Brian mentioned Hard Disk Drive technology which really started with the IBM 1301 in 1961. Although the rate of improvement has been similar to integrated circuits, I would counter that HDD technology is amazing, but only good for memory storage…it doesn’t have the transistor’s breadth of applications.
Both Paul and Brian also said that DNA sequencing has become faster at a rate even greater than Moore’s Law: sequencing a human genome took 13 years in 2003 and cost $3.8 billion, and as of 2012 it could be done in a day and for $1,000. That is indeed “better than Moore’s” but 1) it hasn’t been 50 years of doubling; and 2) while I think DNA sequencing will one day be more transformative than the PC or the internet or LCD screens, that hasn’t happened yet. I have been following and investing in the space since 1996 (I was a VC and on the board of a genomics company in 1998) and the tangible benefits from faster sequencing are lower than most have been hoping for.
But thanks to all those who are actually reading my blog posts and taking the time to try to make them more accurate!