Monday, April 04, 2005

Semiconductor visions by Eli Yablonovitch

Eli Yablonovitch (UCLA) just completed his first Loeb lecture at Harvard, and it's been pretty fascinating. The lecture was somewhat similar to the lecture of the Samsung Electronics' CEO Dr. Hwang, but it was much more informative on the physics issues. (Also, we have not received any USB flash drives this time.)

Yablonovitch (Yabloko is a Slavic word for the "Apple", but no links with Steve Jobs!) is one of the people who are associated with the first chips and microprocessors that have been built, but he has quite a lot to say about the present and the future, too.

His talk has had many dimensions: sociological ones, economical ones, and a great deal of physics. The chip industry has many levels that includes the semiconductor companies; chips; software; hardware; digital content - the broadest categories produce about 3-4 trillion dollars per year.

A simple model of a transistor

Yablonovitch focused on the interplay between physics and economy, but he also presented a very nice simplified caricature of a transistor. It has a gate that can store the electric charge. If there is some electric charge on it, it repels the electrons from a nearby wire, so that the resistance between the "source" and the "sink" increases a lot. With this picture in mind, you can see that it is not difficult for various memory chips to store information for a year. It's simply about storing a small amount of electric charge. Also, if you look at the transistors in this way, you might almost be surprised why you need any semiconductors in the first place. ;-)

The exponential laws

Yablonovitch discussed various exponential laws, especially Moore's law. Things are getting smaller (he cited Feynman as the originator of various visions - "there is a lot of room at the bottom", for example). They're getting faster. We're able to produce more of them - the planet is producing 10^18 transistors per year and this number will soon approach the Avogadro's number. ;-)

An order of magnitude is added every few years which could suggest that the industry is an incredible opportunity for the investors. Unfortunately for the industry, the transistors are also getting cheaper. How does the product, namely the total profit, depends on the year? The total profit (and revenue) has also increased since the 1960s, roughly by a more modest factor of 500. He was remembering his high school project for which he needed to buy some transistors for $10 (a very cheap brand which was not good enough for NASA). Since these old days, the price of a single transistor has decreased by the same ratio that would make a Rolls-Royce from the 1960s cost one nickel today!

Gadgets for tomorrow

Agriculture has become so efficient that only a few percent of the people are working in it today - as opposed to 99% who were farmers thousands of years ago. Will the electronics industry have a similar fate? Is it gonna shrink?

Eli Yablonovitch explained there is a difference: the amount of food that we eat has increased by one order of magnitude per capita at most. We can't eat much more, and therefore as agriculture became much more efficient, the importance of the industry shrank. On the other hand, we expect our electronical devices to perform increasingly more types of functions which potentially allows for the growth of this industry even if its efficiency increases. What are some of his predictions about the future everyday technologies?
  • The first is a mere postdiction today: everything will get wireless. Alex, who was sitting behind me and looking at my blog (greetings!) through his wireless laptop connection, will probably agree. It's already becoming a reality.
  • This includes another trend, also mentioned by Samsung's Hwang, that our cell phones will replace laptops. They will contain everything we need, and if you're gonna need a big screen, you connect your cell phone to the TV set in your hotel (using a wireless connection, of course). Every child, except for the children of irresponsible parents, will get a cell phone at the age of five, a device with 5 buttons (mom, dad, school, police, bathroom or a similar combination).
  • Electronic IDs. Yablonovitch predicts that in a couple of years, everything that costs at least one dollar will have an electronic ID tag on it and an IP address. It's a pretty nice guess about the future. Many things about the privacy can get very subtle or even messy. (Have you also heard about the smart dust?)
  • Translators. He believes that the speech recognition and the electronic translators will become so good that they will be used everywhere.
Limits for tomorrow

Yablonovitch argued that we're approaching the end of the great contribution of George W. Bush to the semiconductor industry: namely the roadmap. What is the final state of the affairs going to look like?
  • Storing information and logical operations are not gonna be hard
  • The main problem is communication as it can take a lot of energy
  • One logical operation currently creates roughly 100,000 kT of energy (where "k" is Boltzmann's constant and "T" is the temperature), and this will decrease to a number as low as 40 kT as the technology becomes nearly optimized
  • The size of the tiniest features of the chip is gonna decrease to the molecular level - roughly 5-10 nanometers; one industrial secret that you can only learn on this blog is that Intel is gonna produce a huge amount of chips based on their new 65-nanometer technology this September instead of the current 90-nanometer technology - so you can see that we're not terribly far from the end
After this era of progress is completed, all transistors have a molecular size and only produce 40 kT of heat per operation. What kind of new progress in design is going to follow this "end of history"? When the current Moore's law stops, what will be the new quantities that may start to grow according to a new law? One of the things he argued is that there will be many microprocessors on one chip which will further speed up many algorithms - a large fraction of algorithms admits parallel computing. And if you don't need the whole field of parallel microprocessors, you just turn most of them off - the energy consumption will continue to matter, he thinks.

Illusion of Rayleigh's laws

The chips are being printed much like the newspapers. In fact, one transistor that used to cost $10 fourty years ago costs as much as one letter in The New York Times. The technology of lithography is very similar to printing newspapers and it's already very cheap.

One of the detailed physical comments was about the validity of Rayleigh's laws. A decade or two ago, many smart and educated people believed that the elementary properties of waves in classical electromagnetism would prevent the people from printing finer sillicon chips than the wavelength of the light that you use. This is why billions of dollars were spent to create X-ray lithography (or, because people usually hate X-rays, it's usually called the deeply ultraviolet lithography, not to be confused with the Planck scale), and so forth. But if you look at the graphs showing the wavelength of light as well as the size of the transistors, the intersection looks as a completely ordinary place in the history of the industry. No one needed the X-rays for creating better chips. It's not funded anymore, and it has become an example of the fact that even if you know the underlying laws well, you can still make incorrect technological decisions.

Around 1999, the transistors within the chips became smaller than the wavelength of the light that you use to create them. How was it possible? Well, they simply overexpose the "film", and this implies that only very tiny "wires" on the chip are left - they're much more tiny than the wavelength of the light. That's great. For a little while, you might think that Rayleigh's laws will continue to hold, albeit with a slightly modified interpretation. The distance between the wires should not be smaller than the wavelength of the light.

But even this interpretation may be incorrect because you can try to print the circuits with multiple expositions. All these considerations, together with the large amount of smart people in the industry, lead Yablonovitch to believe that there will be nothing that can prevent the circuits from becoming as fine as individual molecules. What will be needed, however, is the ability of the engineers and physicists to solve the "inverse problem" - how to combine the various technological steps so that the resulting circuit looks exactly the way we want it to look like.

Moral aspects of the possible collapse of the industry

Yablonovitch is also a co-father of various ambitious companies in the industry, so not too surprisingly ;-), one of his main nightmares is that the revenues in the industry will stop growing in a couple of years. A new kind of bubble may collapse. Many people in the audience disagreed that the collapse of this industry, including millions of highly demanding jobs, is one of the biggest problem facing humanity, and they argued that the humankind has already survived many "revolutions" of this kind in which the percentage of jobs in one well-defined "industry" dropped by a significant margin.

Nevertheless, according to Yablonovitch, such a collapse or slowdown of the chip industry could have far-reaching negative consequences for the whole human society.

A student-activist argued that this progress only affects the developed nations. Yablonovitch disagreed and he explained that in a couple of years, there will be 2 billions of cell phones sold every year. This inevitably implies that many poor people will have to buy the cell phones, too. The proliferation of cell phones is already apparent in villages in China and India, among other places.

Yablonovitch has certainly thought about the future of his industry in detail. He knows a lot and his talk was very inspiring.


  1. Not everything can become wireless. Actually the majority of things can NOT become wireless. Wireless base stations themselves, for example, still must be wired.

    There are two reasons why most things can NOT be wireless:
    1.transportation of energy
    2.transportation of information

    Well there are possible ways of doing both of the above two things wirelessly. But there are NOT efficient way of doing either wirelessly.

    There are ways of transporting energy wirelessly, but to get energy into a device efficiently, you still need to hook a device to electric power lines to either supply energy continuously, or to replenish an energy storage device (rechargeable battery). So you need wires for obtaining energy.

    As for information, the most efficient way of exchange information is still through wires, be it a computer network cable, or even optical fiber. Wireless simply is hard stretched to provide the kind of bandwidth future application requires.

    The Mars Path Finder took 20 minutes to download just one digital image to earth. See how slow that speed is? Because it's wireless! The limited availability of energy (through solar panels) and physics laws prevent the engineering from providing any faster speed. Had it been possible to provide a wired solution to pathfinder, NASA must have loved to prefer the wired over wireless solution.

    Similarly, all wireless devices are bounded by the same thing. They are bounded by battery power. The physics laws dictate it cost certain amount of energy to send certain amount of information to a certain distance away, wirelessly. So the amount of information you can transmit, and the distance you can transmit them, is bounded by the available energy, which is bounded by the battery power, which is bounded by the physical weight of portable devices as well as the chemistry that limits the energy density.

    All those are bounded by physics and we are not too far away from the physical limit on wireless devices. All engineers who ever worked on wireless devices have had to face the headache of battery lifetime issue.

    On the other hand, wired devices do much better because the energy supply is more abundant, plus the energy cost per bit of information is much smaller because all energy largely concentrates on the wires, instead of dissipate into the space in inverse square laws.

    So I actually believe wired solution is much more plausible in a future digital world. It will not take long before we see the availability of computer network jacks every where for a device to be plugged in, just as common place as water fausets and restrooms are every where.


  2. Very much a linguistic aside, Lubos. Anthony Burgess's 'A Clockwork Orange' presents 'yarblockoes' as as a synomyn, nay, a neologism, for the male gonads. I can now see from where he borrowed this term. Recognising that the 'nadsat' lingua franca (and one could not get more lexicograhically eclectic than that) of Alex and the Brotherhood of Saints Cyril and Methodius did so much to inform seventies youth culture ('hey droogie, don't crash here '- D. Bowie), one, who would have defended you from 'furious' and his most horrid accomplices, can only relish the pain of rejection. Not that I'm queer or anything.

  3. Debt Consolidation Debt Consolidation Programs can help you reduce your interest burden by charging an interest rate lower than the rate on your existing loans. Debt consolidation loan can also allow you to make small monthly payments by extending the loan period

  4. Debt consolidation is the best available technique to pay off all past economical debt and economical obligations. In this technique mortgage consolidation of economical debts are done by taking a mortgage consolidation economical debt mortgage.