Monday, March 25, 2019

Quantum Computers Are Analog Computers


Today's topic may be a little afield of conventional engineering ethics, but it involves billions of dollars at risk and the future of an entire engineering-intensive industry, so that's enough to make it an ethical concern already. 

Most engineers have heard of Moore's Law, the observation that eventually became the backbone of the semiconductor industry's road map or marching orders:  the doubling of computing power every two years.  In recent years, Moore's Law has run into difficulties because you simply can't make conventional transistor logic gates too small, or else the electrons don't stay where you want them to because of their quantum nature. 

But not to worry:  for close to four decades now, we've been told that when conventional computer logic circuits can no longer be improved, the industry will switch to "quantum computers," which are based on an entirely different principle that takes advantage of quantum effects, and Moore's Law or its quantum equivalent will keep advancing computer power indefinitely into the future.  This transition to quantum computing has been held out as the best hope for continued progress, and currently it's taken quite seriously by major players in hardware, software, and finance.  IBM and Microsoft, among others, are spending tons of money on quantum computing, and each year  thousands of research papers (mostly theoretical ones) are published about it.

In the face of all this optimism comes one Mikhail Dyakunov, a Russian-born physicist currently at the Université-Montpelier-CNRS in France.  Dyakunov is well-known for his discoveries in plasma and quantum physics over a long career (he is 78).  And last November, the website of the professional engineering magazine IEEE Spectrum published his article "The Case Against Quantum Computing," in which he expresses serious doubts that a practical quantum computer capable of anything more than what conventional computers can do now will ever be built.

Along the way, he gives the most straightforward non-technical explanation of what a quantum computer is that I have seen, and I've seen many over the years.  The gist of the difficulty, he says, is that conventional computers store information as transistor states which are either on or off.  With a clear definition of on and off in terms of a current, say, it's not that challenging to set up and process data in the form of on-or-off bits, which are the essence of what we mean by "digital."  Discrete unambiguous states are the key to the entire conventional-computer intellectual construct, and while errors do occur, there are well-known and not terribly demanding ways to correct them.  That is how we got to where we are today.

But the fundamental logical unit in a quantum computer is not a conventional on-or-off current or voltage.  It is the quantum state of a "qubit" which can be exemplified by, for instance, the direction that the magnetic axis of an electron points in.  And as long as you are not taking a measurement (roughly equivalent to reading out data), the information that makes quantum computing work is the exact angle of that spin with respect to some reference direction.  And that angle is not just up or down, 1 or 0, but can take on any value between plus and minus 90 degrees. 

Back where I come from, a computer which stores information in the form of continuous physical states is called an analog computer.  Most people younger than 40 have little or no memory of analog computers, but surprisingly sophisticated problems were solved on these things from the early 20th century up to the 1960s.  However, they were comparatively slow and had very limited accuracy, typically a percent or so.  And when digital computers came along, virtually all analog computers became museum pieces (think of how many people you see using slide rules these days).  One of the last ones to go was a curious system that took synthetic-aperture radar (SAR) data from a flying airplane and transformed the data into light and dark patches on photographic film.  Then the film was placed into an optical system that performed a Fourier transform on the data and presto!  you obtained the real-space version of the SAR radar image:  the actual mountains and valleys that the plane flew over.  Since this gizmo used light waves, and light waves are fundamentally quantum in nature, I suppose you could have called that a quantum computer, though nobody did.

And you can bet nobody who is promoting quantum computing is going to refer to their goal as an analog computer, because for decades, "analog" has been an embarrassing term in the world of computation.  But guess what—Dyakunov has explained to us mortals that quantum computers have to manipulate and store data in analog form.  And the same kinds of problems of accuracy and errors that caused the analog-computer dinosaurs to die off are currently keeping quantum computers from getting any farther than they have so far, which is not very (no practical quantum computers are in commercial production).  You think reading out a analog computer's shaft position accurately is hard?  Try measuring the spin of a single electron without disturbing it.  I may be oversimplifying things, ut that seems to be the essence of what has to be done.  And Dyakunov points out that the experts themselves say they'll need thousands of logical qubits to do anything useful, and perhaps up to a thousand physical qubits per logical qubit to have enough information to correct the inevitable errors. 

In sum, Dyakunov thinks the quantum-computing fad may be going the way of the superconducting-computer fad, which flared in the 1980s and died in the early 2000s when conventional silicon-based computers overtook them performance-wise.  For a time, it was easier to build smaller logic gates out of something called Josephson junctions than it was to make silicon gates.  The problem with Josephson junctions is that they have to be cooled to a few millikelvin with liquid helium, which leads to all kinds of interface problems.  Ironically, Josephson junctions are one of the leading contenders for the best path to qubits, but handling millikelvin circuits hasn't gotten much easier in the meantime. 

The late science-fiction writer Arthur C. Clarke made a famous comment about elderly scientists: "When a distinguished but elderly scientist states that something is possible, he is almost certainly right. When he states that something is impossible, he is very probably wrong."  By this criterion, we should ignore Dyakunov and keep working on quantum computers.  But it would be interesting if he turns out to be right. 

Sources:  I read Dyakunov's article in the March 2019 hard-copy issue of IEEE Spectrum, pp. 24-29, but a version is also available online at https://spectrum.ieee.org/computing/hardware/the-case-against-quantum-computing.  I also referred to https://prabook.com/web/mikhail.dyakonov/448309 for Dyakunov's date of birth and the Wikipedia article on him, and the Arthur C. Clarke entry in Wikiquote for what is known as Clarke's First Law. 
-->

No comments:

Post a Comment