Karl Skjonnemand – TED Talk TRANSCRIPT
Computers used to be as big as a room. But now they fit in your pocket, on your wrist and can even be implanted inside of your body.
How cool is that? And this has been enabled by the miniaturization of transistors, which are the tiny switches in the circuits at the heart of our computers. And it’s been achieved through decades of development and breakthroughs in science and engineering and of billions of dollars of investment.
But it’s given us vast amounts of computing, huge amounts of memory and the digital revolution that we all experience and enjoy today.
But the bad news is, we’re about to hit a digital roadblock, as the rate of miniaturization of transistors is slowing down. And this is happening at exactly the same time as our innovation in software is continuing relentlessly with artificial intelligence and big data.
And our devices regularly perform facial recognition or augment our reality or even drive cars down our treacherous, chaotic roads. It’s amazing. But if we don’t keep up with the appetite of our software, we could reach a point in the development of our technology where the things that we could do with software could, in fact, be limited by our hardware.
We’ve all experienced the frustration of an old smartphone or tablet grinding slowly to a halt over time under the ever-increasing weight of software updates and new features. And it worked just fine when we bought it not so long ago.
But the hungry software engineers have eaten up all the hardware capacity over time. The semiconductor industry is very well aware of this and is working on all sorts of creative solutions, such as going beyond transistors to quantum computing or even working with transistors in alternative architectures such as neural networks to make more robust and efficient circuits.
SLOWDOWN OF MINIATURIZATION
But these approaches will take quite some time, and we’re really looking for a much more immediate solution to this problem. The reason why the rate of miniaturization of transistors is slowing down is due to the ever-increasing complexity of the manufacturing process. The transistor used to be a big, bulky device, until the invent of the integrated circuit based on pure crystalline silicon wafers.
And after 50 years of continuous development, we can now achieve transistor features dimensions down to 10 nanometers. You can fit more than a billion transistors in a single square millimeter of silicon.
And to put this into perspective: a human hair is 100 microns across. A red blood cell, which is essentially invisible, is eight microns across, and you can place 12 across the width of a human hair. But a transistor, in comparison, is much smaller, at a tiny fraction of a micron across.
You could place more than 260 transistors across a single red blood cell or more than 3,000 across the width of a human hair. It really is incredible nanotechnology in your pocket right now.
And besides the obvious benefit of being able to place more, smaller transistors on a chip, smaller transistors are faster switches, and smaller transistors are also more efficient switches. So this combination has given us lower cost, higher performance and higher efficiency electronics that we all enjoy today.
To manufacture these integrated circuits, the transistors are built up layer by layer, on a pure crystalline silicon wafer. And in an oversimplified sense, every tiny feature of the circuit is projected onto the surface of the silicon wafer and recorded in a light-sensitive material and then etched through the light-sensitive material to leave the pattern in the underlying layers.
And this process has been dramatically improved over the years to give the electronics performance we have today.
But as the transistor features get smaller and smaller, we’re really approaching the physical limitations of this manufacturing technique. The latest systems for doing this patterning have become so complex that they reportedly cost more than 100 million dollars each. And semiconductor factories contain dozens of these machines.
So people are seriously questioning: Is this approach long-term viable? But we believe we can do this chip manufacturing in a totally different and much more cost-effective way using molecular engineering and mimicking nature down at the nanoscale dimensions of our transistors.
As I said, the conventional manufacturing takes every tiny feature of the circuit and projects it onto the silicon. But if you look at the structure of an integrated circuit, the transistor arrays, many of the features are repeated millions of times. It’s a highly periodic structure. So we want to take advantage of this periodicity in our alternative manufacturing technique.
We want to use self-assembling materials to naturally form the periodic structures that we need for our transistors. We do this with the materials, then the materials do the hard work of the fine patterning, rather than pushing the projection technology to its limits and beyond.
Self-assembly is seen in nature in many different places, from lipid membranes to cell structures, so we do know it can be a robust solution. If it’s good enough for nature, it should be good enough for us. So we want to take this naturally occurring, robust self-assembly and use it for the manufacturing of our semiconductor technology.
One type of self-assemble material — it’s called a block co-polymer — consists of two polymer chains just a few tens of nanometers in length. But these chains hate each other. They repel each other, very much like oil and water or my teenage son and daughter.
But we cruelly bond them together, creating an inbuilt frustration in the system, as they try to separate from each other. And in the bulk material, there are billions of these, and the similar components try to stick together, and the opposing components try to separate from each other at the same time.
And this has a built-in frustration, a tension in the system. So it moves around, it squirms until a shape is formed. And the natural self-assembled shape that is formed is nanoscale, it’s regular, it’s periodic, and it’s long range, which is exactly what we need for our transistor arrays.
So we can use molecular engineering to design different shapes of different sizes and of different periodicities. So for example, if we take a symmetrical molecule, where the two polymer chains are similar length, the natural self-assembled structure that is formed is a long, meandering line, very much like a fingerprint.
And the width of the fingerprint lines and the distance between them is determined by the lengths of our polymer chains but also the level of built-in frustration in the system. And we can even create more elaborate structures if we use unsymmetrical molecules, where one polymer chain is significantly shorter than the other.
And the self-assembled structure that forms in this case is with the shorter chains forming a tight ball in the middle, and it’s surrounded by the longer, opposing polymer chains, forming a natural cylinder. And the size of this cylinder and the distance between the cylinders, the periodicity, is again determined by how long we make the polymer chains and the level of built-in frustration.
So in other words, we’re using molecular engineering to self-assemble nanoscale structures that can be lines or cylinders the size and periodicity of our design.
We’re using chemistry, chemical engineering, to manufacture the nanoscale features that we need for our transistors. But the ability to self-assemble these structures only takes us half of the way, because we still need to position these structures where we want the transistors in the integrated circuit.
But we can do this relatively easily using wide guide structures that pin down the self-assembled structures, anchoring them in place and forcing the rest of the self-assembled structures to lie parallel, aligned with our guide structure.
For example, if we want to make a fine, 40-nanometer line, which is very difficult to manufacture with conventional projection technology, we can manufacture a 120-nanometer guide structure with normal projection technology, and this structure will align three of the 40-nanometer lines in between. So the materials are doing the most difficult fine patterning.
And we call this whole approach “directed self-assembly.” The challenge with directed self-assembly is that the whole system needs to align almost perfectly, because any tiny defect in the structure could cause a transistor failure.
And because there are billions of transistors in our circuit, we need an almost molecularly perfect system. But we’re going to extraordinary measures to achieve this, from the cleanliness of our chemistry to the careful processing of these materials in the semiconductor factory to remove even the smallest nanoscopic defects.
So directed self-assembly is an exciting new disruptive technology, but it is still in the development stage.
But we’re growing in confidence that we could, in fact, introduce it to the semiconductor industry as a revolutionary new manufacturing process in just the next few years. And if we can do this, if we’re successful, we’ll be able to continue with the cost-effective miniaturization of transistors, continue with the spectacular expansion of computing and the digital revolution.
And what’s more, this could even be the dawn of a new era of molecular manufacturing. How cool is that?