Featured Article : Have Computers Reached Their Limit?
Many tech commentators have noted how the stagnation in computing has led to ‘Moore’s Law’ being challenged, but has the shrinking of transistors within computer chips really hit a wall and what could be next for Moore’s Law?
What Is Moore’s Law?
Moore’s Law, named after Intel co-founder Gordon Moore, is based on his observation from 1965 that transistors were shrinking so quickly that twice as many would be able to fit into a micro-chip every year, which he later amended to a doubling every two years. In essence, this Law should mean that processing power for computers doubles every two years.
The Big Problem
The big problem is that technical challenges to Moore’s Law have led to a slow-down period that now appears to be challenging the validity of the Law itself. For example, the growth of the Internet and the IoT, plus mass digitisation and a reliance upon technology in the developed world all mean that economic growth within industry, science, medicine and more now rely upon computers/devices and connections to be faster, cheaper, and more widely available. Limitations in chip manufacture and just how much improvement can now be made in terms of performance and power consumption and fitted into ever-smaller spaces in chips could, therefore, be a huge threat to progress and innovation in many aspects of life.
The challenges to Moore’s Law that many tech commentators have noted are that:
– Technology companies may be reaching their limit in terms of fitting ever-smaller silicon transistors into ever-smaller spaces, thereby leading to a general slowing of the growth of processing power.
– Big computer chip manufacturers like Intel have delayed the next generation of smaller transistor technology and increased the time between introducing the future generations of their chips. Back in 2016 for example, Intel found that it could shrink chips to as little as 14 nanometres, but 10 nanometres is going to be a challenge that would take longer to achieve. Intel, for example, is taking five years, rather than two, to make its latest ‘process node transition’.
– With chip complexity doubling every other year, the scalability of new chips design innovations is a challenge in itself.
– Chip manufacture has been disrupted and there is now a global chip shortage which is having huge knock-on-effects across many industries e.g., car plant shutdowns due to chip shortages.
The knock-on effects of these challenges to Moore’s Law are that:
– There now appears to be a slowing of computer innovation that some say could have a detrimental effect on new, growing industry sectors such as self-driving cars.
– Big tech companies are finding it more difficult to keep improving their data centres.
– The rate of improvement of supercomputers has been slowing in recent years and this may have had a negative impact on the research programs that use them.
– Computers are being challenged in how they can work for (and keep up with) the demands of business.
– Mobile devices, which use chips other than Intel’s, may also have the brakes put on them slightly as they now also rely, to a large extent, on the data-centres to run the apps that their users value.
The target for chip manufacturers is finding new ways to reach staggeringly small, 2nm chips that can still deliver performance and are energy efficient, in large enough numbers to meet world demand by 2026.
Some of the possible ways forward and technical solutions being worked upon and introduced to prolong the life of Moore’s Law include:
– Changing transistor designs e.g., to Gate All Around (GAA), also known as nanoribbon or nanosheets (horizontally stacked nanosheets). This extension of the FinFET concept produces transistors that look like cylinders with a gate coat all around, thereby getting around challenges like variability and mobility loss.
– New storage technologies such as Spin Transfer Torque (STT-RAM), and new logic technologies such as transistor lasers, Domain Wall Logic, or Spin Wave, could help to prolong the life of Moore’s Law, but may not be ready in time.
– Using numeric analysis, graph analysis, and spectral graph theory techniques. For example, researchers at Stevens Institute of Technology have used numerical matrices (known as eigenvalues and eigenvectors) to develop algorithms that make it easier to understand the relationships between billions of different chip components. This, in turn, has enabled the development of software tools such as Graph Spectral Sparsifier (GRASS) which chip designers can use to simplify the chip design process so that an integrated circuit with billions of elements to be analysed in just a few hours.
– Using atomic or molecular-level etching and deposition to more precisely target and treat areas of chips, thereby improving yield and throughput in chip manufacture.
– Using liquid cooling in data centres. For example, Microsoft has recognised that it has now come up against the slowdown of Moore’s Law as transistor widths have shrunk to atomic scales and are reaching a physical limit, whilst the demand for faster computer processors for high performance applications such as AI has accelerated. This has meant that more electric power is now being put through the small processors used in Microsoft’s data centres, thereby increasing the heat they produce. According to Microsoft, this means that air cooling is no longer enough to prevent the chips from malfunctioning. The demands of a huge increase in the numbers of Teams users during lockdown and the need to maintain sustainable and energy efficient data centres have also contributed to Microsoft’s decision to try liquid cooling. Microsoft has, therefore, adopted a new system of two-phase immersion cooling which involves immersing servers in tanks filled with an engineered fluid (from 3M) which has dielectric properties (i.e., it is an effective insulator), thereby allowing the servers to operate normally while fully immersed in the fluid. The liquid boils at 122 degrees Fahrenheit (90 degrees lower than the boiling point of water) and this boiling effect, generated by the work the servers are doing, takes the heat away from the computer processors whilst the low-temperature boil enables the servers to operate continuously at full power without risk of failure due to overheating. The second phase of this two-phase process refers to the vapour rising from the tanks making contact with a cooled condenser in the tank lid, thereby changing it back to liquid that rains back onto the immersed servers, creating a closed-loop cooling system. The result is the ability to continue the Moore’s Law trend at datacentre level as well as reducing power consumption.
– AI designing chips. For example, a recent Google research paper has described how a deep reinforcement learning approach to chip ‘floorplanning’ has led to AI generating chip floorplans that are superior or comparable to those produced by humans. The researchers used a deep reinforcement learning approach, coupled with an edge-based graph convolutional neural network architecture to design the next generation of Google’s artificial intelligence (AI) accelerators, and this method enabled the AI to learn from past experience in chip floorplanning and to become better and faster at solving new instances of the problem. The researchers found that in just under six hours, instead of the months it would have taken for human engineers, the AI design method automatically generated chip floorplans that are superior or comparable to those produced by humans in all key metrics, such as power consumption, performance, and chip area. Also, the researchers believe that in addition to AI’s success in designing chips for AI, more powerful AI-designed hardware is likely to fuel advances in AI itself, thereby creating a kind of symbiotic relationship between the two fields.
What Does This Mean For Your Business?
Many smaller businesses that are less directly reliant upon the most-up-to-date computers may not be particularly concerned at the present time about the challenge to Moore’s Law, but all businesses are likely to be indirectly affected as their tech giant suppliers struggle to keep improving the capacity of their data-centres, and chip manufacturers struggle with the challenges of chip shortages coupled with the technical difficulties of designing and creating smaller chips with the right levels of performance fast enough.
Many see AI and machine learning as the gateway to finding innovative solutions to improving chip design and computing power, but these also rely on data centres and other areas of computing that have been challenged by the pressure on Moore’s Law.
A more likely way forward might be that chip designs will need to be improved and highly specialised versions will need to be produced, and Microsoft and Intel have already made a start on this by working on reconfigurable chips. Also, the big tech companies may need to collaborate on their R &D in order to find the way forward in increasing the rate of improvement of computing power that can ensure that businesses can drive their products, services, and innovation forward.