It’s tough to find a word that adequately describes the impact of computers on how we work, communicate, learn, and travel. The rate of progress has been equally staggering – there is more compute power in our smartphones than in the rocket that sent Neil Armstrong to the Moon in 1969, all enabled by incredible innovation in the design and fabrication of silicon transistor chips, the bedrock of classical compute.

But this rate of progress – the doubling of processing speed or power every two years or so - is in jeopardy . These improvements are driven by reducing the size of the chip features, but achieving ever smaller geometries is increasingly challenging, expensive, and time-consuming.

Of course, not all products require the general-purpose performance capabilities of a CPU chip, and limitations of CPU chips can be addressed by using a custom chip that optimises a particular performance metric. AI chips that have been designed to achieve high-speed AI processing are one example of this. However, even for these chips, it can take months just to get a prototype from a foundry, slowing your innovation and increasing the time to market. And if you need only low volumes, the high initial development cost might also mean increased cost per unit.

And lastly there are some problems that are simply intractable with classical digital compute based on bits and bytes. Add to all this the increasing concerns about the sustainability of computing applications as the world creates, processes and stores ever more data and it is obvious that we need a new approach to computing.

The good news is that all these factors are driving innovation in numerous new computing technologies. The not-so-good news is that none of them will provide a one-size-fits-all solution, so it will be critical for businesses to understand exactly what they need.

Given the ubiquity, usefulness, and relatively low cost of our existing silicon CPU technologies, it is not surprising that a more effective drop-in replacement has not yet been identified. Newer computing technologies are being developed to fill a particular niche in terms of capability and, as a result, will almost certainly be used as part of a hybrid system that combines them with classical silicon chip computing. Further, these new technologies have different time horizons to maturity (see Figure below). Based on what we’ve seen to date, there are three broad drivers for selecting novel compute technologies and they roughly correlate with time to maturity.

a) Pushing the bounds of silicon

There is a strong argument for trying to get more out of silicon, given the decades of research and development to date. Silicon is plentiful and inexpensive compared to other semiconductor materials; we understand its properties and have facilities to fabricate and package devices in large volumes; and we have practical user interfaces, operating systems, and programming languages. Silicon also offers a straightforward route for integration of components, a major driver of research into silicon photonics, for example. Silicon offers some of the nearer-term future compute technologies. Novel Architectures (described in Section 3.5) started to be a reality one or two years ago, while neuromorphic computing and photonic computing, both based on silicon, are likely to be commercially available to the mass market in five or six years.

b) Achieving a step change in performance, cost or power consumption using a novel approach

Our computing performance has come a long way since the first transistor – we have seen a trillion-fold increase of computational speed measured in Floating Operations Per Second (FLOPS) over the past 60 years, while the cost of silicon chips has decreased by a factor of 20%-30% annually. So, what do we mean by a step change? Alternative materials systems and approaches could leapfrog our current capabilities with silicon:

- All-optical photonic computing could offer an increase in processing speed of 10 or 100 times
- Writing information into DNA offers the ability to store vast amounts of data very inexpensively in a smaller and passive form, thereby reducing energy cost in comparison to traditional silicon.
- And neuromorphic computers using novel, even quantum, materials could enable much more compact, low power devices that behave more like a human brain

c) Making the impossible possible

Despite the great advances in classical computing, some computational problems are still intractable. For example, the Travelling Salesman Problem is a classical problem that tries to find the optimal route between two locations, stopping only once at every intermediate point required. It can be easily solved for a few intermediate points, but the computational time required grows exponentially as more locations need to be considered. This is a daily problem that supply chain and logistic operations face that would currently take a classical computer thousands of years to solve, even without taking into account external factors such as traffic jams, diversions, and delivery times.

Quantum computing has gained much attention lately because it offers a way to overcome the limitations of digital bits, thereby paving the way for us to simulate and explore complex physical systems, such as large molecules, in a way that is not currently possible even with the most powerful supercomputers. Since simulations can be run in a fraction of the time of physical trials, quantum computing offers the possibility to reduce development time for drugs, for example, or accelerate discovery of new materials.

With the physical limitations of reducing the size of foundational electronic components (i.e., transistors and diodes), the future of silicon-based electronics will be dependent on layered designs. So far, these basic components have been manufactured in one-layer wafers, but now multi-layer designs are being considered to keep producing even quicker and smaller chips for a lower price. These computational improvements aim, for instance, to allow personalised chip designs that cope with AI algorithms needs and provide enhanced experience for users.

Neuromorphic computing does not follow a sequential processing as traditional computers do, instead they try to imitate how neurons process information. Our neurons are stimulated through spikes that can trigger one or multiple neurons, which can then activate others they are connected with. This means that, like neurons, neuromorphic computing will be capable of parallel and targeted processing for a define set of neurons and interconnections. Its potential will be explored, for instance, by algorithms and sensing systems that try to imitate the human thinking.

The vast majority of computing today makes use of the energy contained in electrons to process data. The fundamental idea of photonic computing is to replace the electrons in digital computers with photons, using light waves to process and store data instead. Due to the speed of light being unsurpassable, photonics provides a theoretical minimum latency and therefore makes light a better computing medium compared to electricity. In principle, we could also observe a 10-50x the bandwidth improvement over traditional computing due to the bits travelling at the speed of light. Furthermore, a 10x energy efficiency can be obtained with optical computing as one can increase the processing power without having to increase the electricity supply, which is the case for classical electronic computers.

Biological computing uses biologically derived molecules such as DNA and proteins to perform computation, instead of the electrical signals used in today’s digital computers. As a result of advances in nanobiotechnology, researchers have discovered methods of programming living cells to respond in predictable ways to certain chemical inputs. While programming such cells is costly and labour intensive in the first instance, it becomes extremely cost-effective to grow billions more once this is achieved. Biological computing has already been successfully used for cold data storage by encoding information in DNA but will eventually be able to process data with very high reliability. 

Quantum computing is based on the physical phenomena where electrons can simultaneously behave as material particles or waves. This property changes the basic idea of storing and processing information in the form of binary bits (0 or 1) and instead using qubits (any number from 0 to 1). Particles will be able to represent any of an infinite number of states and interact with other particles in infinite different ways. Therefore, quantum computers have the potential of solving very complex problems faster than traditional computers.