Friday, May 22, 2015

Death of Moore's Law? Don't count on it

Where a calculator on the ENIAC is equipped with 18,000 vacuum tubes and weighs 30 tons, computers in the future may have only 1,000 vacuum tubes and perhaps weigh 1.5 tons.Popular Mechanics, March 1949

Integrated circuits will lead to such wonders as home computers -- or at least terminals connected to a central computer -- automatic controls for automobiles, and personal portable communications equipment. Gordon Moore, 1965

For centuries, explorers have searched the world for the fountain of youth. Today’s billionaires believe they can create it, using technology and data. -- Ariana Eunjung Cha, April 4, 2015

The phrase “hit a wall” in idiomatic English means to reach a point where progress stops or slows significantly.  Engineers don’t warm up to that phrase very easily.  For an engineer a wall is a challenge, more like a speed bump. They throw their hands up only if they’re being robbed – maybe.  They have little use for the word “impossible.”

Let’s say you’ve written a book on genetics, and as a way of getting would-be readers pumped up you decide to store every word and illustration in your book – table of contents, index, everything -- in DNA.  And not only store it, but retrieve it as well.  And to get your point across about the immenseness of DNA storage someone suggests you write 70,000,000,000 (70 billion) copies of the book in DNA.  Impossible?  Of course not.  Geneticist George Church did it three years ago. 

On April 19, 1965 an engineer named Gordon Moore published an article in which he noted something remarkable about integrated circuits and their components.  Often referred to as a chip, an integrated circuit is a microelectronic device consisting of many interconnected transistors and other components fabricated on a semiconductor wafer, usually silicon.  Without integrated circuits we would have no smartphones, tablets, PCs, Macs, and countless other electronic devices.  “For simple circuits,” Moore wrote, “the cost per component is nearly inversely proportional to the number of components.”

This was the engineering equivalent of striking gold.  With more components not only did you get increased performance (since electrons have a shorter commute), but the cost per component decreased -- at least up to a point.  At the time, chips contained only a handful of components, yet Moore predicted an exponential trend was underway, so that by 1975 “the number of components per integrated circuit for minimum cost will be 65,000.”

This was a bold extrapolation on Moore’s part.  Prior to using integrated circuits transistors and other components were wired together by hand on a circuit board.  A painstaking process, but one that, early on, was less expensive than producing integrated chips.  In 1965 only a few companies were making integrated circuits, and their customers were mostly NASA and the U.S. military.  Add to this the fact that only about 10-20 percent of the transistors actually worked, according to Moore’s recollection.

Yet by 1975 Intel, the company he co-founded in 1968, was preparing to market a memory chip with 32,000 components, leaving Moore’s original estimate off only by a factor of two. 

This progression of a doubling of transistor density every year, later revised to every two years, became known as Moore’s Law.  Higher densities meant increased performance, more useful technology.  Higher densities drove component cost down, deflating retail prices and attracting more consumers.

Bolstered by other technological innovations, 50 years of Moore’s Law has brought us the Digital Age, as Moore predicted.  More precisely, engineers, project managers, and entrepreneurs have innovated, invested and generally dedicated their lives to keep Moore’s Law going.  And consumers eagerly line up for the latest offerings.

How far have we come since Moore wrote his article?  Today’s integrated circuits contain billions of transistors and are fabricated in factories that run in the billions of dollars to build.  Yet the cost per transistor has dropped from about $30 in 1965 (in 2015 dollars) to an infinitesimal amount today.

And the world has an insatiable hunger for transistors.  As one researcher noted,

In 2014, semiconductor production facilities made some 250 billion billion (250 x 10^18) transistors. This was, literally, production on an astronomical scale. Every second of that year, on average, 8 trillion transistors were produced. That figure is about 25 times the number of stars in the Milky Way and some 75 times the number of galaxies in the known universe.

Remember, that’s 8 trillion transistors for every second of 2014.  The flood of transistors has “been the ever-rising tide that has not only lifted all boats but also enabled us to make fantastic and entirely new kinds of boats.”

Though there have been predictions in the past of the imminent demise of Moore’s Law, engineers were able to find ”ways around what we thought were going to be pretty hard stops,” as Moore stated in a recent interview.  But ultimately, there are the fundamental limits of the known world: the speed of light and the atomic nature of materials. 

As Ray Kurzweil has pointed out frequently and which a surprising number of researchers seem to forget, Moore’s Law is a computing paradigm and is the fifth such paradigm since the 1890 census.  All five paradigms have shown exponential growth.  The first four – electromechanical, relay, vacuum tube, and transistor – eventually lost steam and were superseded by a technology previously found only in niche markets, such as the military, or that languished in sparsely-funded research labs.  As a paradigm slows -- no longer progresses exponentially -- more dollars are spent on developing the most promising technologies to replace it.

Going 3D

The semiconductor industry is running out of tricks to keep shrinking silicon transistors.  Right now the smallest size is 14 nanometers, and by 2020 they will need to be five nanometers to keep pace with Moore’s Law.  Is this a wall approaching? 

Not to Samsung.  They’re already building flash memory chips using three-dimensional integrated circuits to achieve performance gains.  Instead of piling transistors side-by-side on a plane of silicon, they’re stacking them, taking up half the space of planar chips.  They’re building hi-rise condos instead of subdivisions with smaller and smaller houses.  More layers means better performance and no more shrinking.  No longer will they have to retrofit multibillion-dollar factories to produce the latest chips. 

IBM is taking a different approach, at least for now. They’re developing transistors built with carbon nanotubes instead of silicon, which they hope to have ready for mass production by 2020. At two nanometers in diameter, the nanotubes, which though seamless resemble rolled up chicken wire, could continue the pace of cramming more transistors onto a silicon substrate.  Based on simulations, the nanotube transistors are about five times as fast as ones made from silicon.

In his magnum opus Kurzweil cites the work of Peter Burke, University of California/Irvine, who demonstrated nanotube circuits operating at 2.5 gigahertz (2.5 GHz).  However, in a peer-reviewed article Burke claimed the theoretical speed limit of these nanotube transistors should be measured in terahertz, where 1 THz equals 1,000 GHz.  (What a boost that would be to my 2.66 GHz MacBook Pro!)

The biggest problem is positioning the nanotubes closely enough together on the chip.  IBM’s preferred approach is to label the substrate and nanotubes with a compound “that would cause them to self-assemble into position.”

Self-assembly of nanoscale circuits would be a world-changer.  As Kurzweil notes, citing the work of researchers at UCLA, having “potentially trillions of circuit components organize themselves, rather than be painstakingly assembled in a top-down process, would enable large-scale circuits to be created in test tubes rather than in multibillion-dollar factories, using chemistry rather than lithography.”

Creating nanocircuits in chemistry flasks “will be another important step in the decentralization of our industrial infrastructure and will maintain the law of accelerating returns through this century and beyond.”

In two decades robots will be in charge

One of the reasons for continuing the exponential development of technology is to eventually turn the task over to robots.  They will eventually graduate and take charge of R&D.  In the 2020s, working with advanced hardware and computational strategies, researchers will make major progress in emulating the human brain.  By 2029 a computer will be able to pass itself off as human under competent interrogation during a Turing Test.

Meanwhile, the inexorable march of miniaturization will make its way into our bodies, including our brains.  Nanobots the size of a red blood cell will enter our bloodstream and augment our intelligence, combining the pattern recognition power of our biological brains with the speed, capacity, and knowledge-sharing ability of our technology.  This should get underway sometime in the 2020s. 

By the 2030s the nonbiological portion of our intelligence will predominate.  Somewhere around 2045 “the pace of technological change will be so rapid, its impact so deep, that human life will be irreversibly transformed.”  Kurzweil calls this period the Singularity.

AI researcher Ben Goertzel thinks the Singularity could arrive much sooner – in 10 years.

Technology will continue to empower us with better and cheaper products some of which will give us the ability to make better and cheaper products. Keynesianism and its Free Lunch Institute known as the welfare state will gradually kill off banks and their governments as we've known them.  Our overlords will not go quietly but there is a better future ahead.

As a hint of that better future I offer this sampling of encouraging developments:

1.    Surgeon Anthony Atala demonstrates an early-stage experiment that could someday solve the organ-donor problem: a 3D printer that uses living cells to output a transplantable kidney. Using similar technology, Dr. Atala's young patient Luke Massella received an engineered bladder 10 years ago; we meet him onstage.

2.    Just like his beloved grandfather, Avi Reichental is a maker of things. The difference is, now he can use 3D printers to make almost anything, out of almost any material. Reichental tours us through the possibilities of 3D printing, for everything from printed candy to highly custom sneakers.

3.    3D printing has grown in sophistication since the late 1970s; TED Fellow Skylar Tibbits is shaping the next development, which he calls 4D printing, where the fourth dimension is time. This emerging technology will allow us to print objects that then reshape themselves or self-assemble over time. Think: a printed cube that folds before your eyes, or a printed pipe able to sense the need to expand or contract.

4.    What we think of as 3D printing, says Joseph DeSimone, is really just 2D printing over and over ... slowly. Onstage at TED2015, he unveils a bold new technique — inspired, yes, by Terminator 2 — that's 25 to 100 times faster, and creates smooth, strong parts. Could it finally help to fulfill the tremendous promise of 3D printing?

Source: The Singularity is Near, p. 66

The chart shows the price-performance of forty-nine computational systems in the 20th century, measured by instructions per second per thousand constant dollars.  (A rising straight line on a log chart indicates exponential growth.)