Will IBM’s 3D chips bring Moore’s Law back to life?

The new process developed by IBM could open the door to the mass production of 3D chips, which in turn could revive Moore’s Law.

At the time of the rise of computing, everyone swore by Moore’s Law popularized by the engineer of the same name. It stipulates that with the advancement of technology, the number of transistors (these small sub-units at the base of the architecture of a computer) in the same integrated circuit is set to double every two years, which results in a rapid increase in power of the machines.

But contrary to what its name suggests, it is not a “law” strictly speaking, but a simple empirical observation. However, the technological landscape has changed considerably since Moore’s prophecy; for several years, we have seen a significant slowdown in this trend.

The causes of this slowdown are still unclear and potentially very numerous. One can in particular cite the dazzling progress of the software side, in particular with the explosion of AI; today, it is possible to achieve very high computing power with less equipment than at the time. Multiplying the number of transistors is therefore no longer as interesting.

A graph that illustrates the slowing down of the trend described by Moore’s Law. © John Hennessy/David Patterson

In addition, manufacturing processes are becoming extremely sophisticated; the closer we get to the physical limits of manufacturing processes, particularly in terms of miniaturization, the more it is difficult to maintain this frantic pace of innovation. We can also mention the pandemic and the shortage of semiconductors which have weighed heavily on supply chains.

Use volume instead of excessive miniaturization

Is Moore’s Law already dead and buried? Not necessarily, according to IBM. The computer giant has teamed up with Japan’s Tokyo Electron to develop a new technique for producing 3D chips; the researchers consider that it could simplify the supply chain to the point of bringing this notion back to the fore.

On a standard chip, the transistors are arranged flat on the same 2D plane. In a 3D chip, they are stacked and connected on top of each other to drastically increase the amount of transistors in a given volume. What go beyond the framework of Moore’s law which reasons in terms of surface, or with one dimension less.

This 3D chip concept is not new. All the giants in the sector, starting with Intel, are already interested in it (see the video above). But for the moment, it is still reserved for extreme and highly specialized applications ; integrators use it, for example, to produce ultra-fast random access memory for supercomputers.

If they are still conspicuous by their absence from the consumer segment, it is because the manufacturing process is very complicated. It’s hard enough to arrange lots of tiny transistors on a surface, but it’s even harder in three dimensions.

A new manufacturing process

Traditionally, the chips are first fixed on a glass support. This reinforces it to ensure that it will survive the manufacturing process. The problem is that at the end of the process, this layer of glass must also be removed using UV lasers.

The process works very well and is already mastered by the industry for standard chips. But the presence of this glass plate still represents a big constraint. This involves many precautions to ensure that the chip has not been damaged; and for now, this approach therefore remains incompatible with large-scale 3D chip production.

The new manufacturing process makes it possible to dispense with glass plates, which theoretically opens the door to the manufacture of large-scale 3D chips. © IBM

The researchers therefore chose another approach. Instead of using a glass support, they bet on a silicon plate. Again, this is a technique that already exists. It has the advantage of being much more practical on an industrial scale. However, it is rarely used. Because according to IBM, the separation process requires the use of a physical force which can introduce many manufacturing defects.

This is where the novelty of the work of IBM and TEL lies. After four years of effort, they succeeded in developing a new module that allows separate the chip from its support using an infrared laser. This does not affect the silicon at all; this allows to preserve the perfect integrity of the chipe, and by extension to unlock the use of this technology to produce large-scale 3D chips.

Heading for the IT of tomorrow

With the technology now firmly in place, the two partners will begin to explore the industrial side of this approach. ” We will show how this process could be integrated into a semiconductor manufacturing line, and thus demonstrate that our process is compatible with a complete 3D chip production line. “, explains the press release.

This does not mean that you will soon be able to replace your computer’s processor with a 3D CPU, far from it. This approach is still very experimental and far from landing on the mainstream segment. In all likelihood, we will still have to wait long years.

But these 3D structures look more and more like the logical and natural evolution of current chips. Each time a company shows significant progress on this issue, we are therefore one step closer to the next great paradigm of computing. It will therefore be interesting to follow the evolution of this technology, which may allow Moore’s law to rise from the ashes.

Leave a Comment