If neuromorphic computing is needed, how can it be achieved? First, the technical requirements. Bringing together diverse research communities is necessary but not sufficient. Incentives, opportunities and infrastructure are needed. The neuromorphic community is a disparate one lacking the focus of quantum computing, or the clear roadmap of the semiconductor industry. Initiatives around the globe are starting to gather the required expertise, and early-stage momentum is building. To foster this, funding is key. Investment in neuromorphic research is nowhere near the scale of that in digital AI or quantum technologies (Box 2). Although that is not surprising given the maturity of digital semiconductor technology, it is a missed opportunity. There are a few examples of medium-scale investment in neuromorphic research and development, such as the IBM AI Hardware Centre’s range of brain-inspired projects (including the TrueNorth chip), Intel’s development of the Loihi processor, and the US Brain Initiative project, but the sums committed are well below what they should be given the promise of the technology to disrupt digital AI.
The neuromorphic community is a large and growing one, but one that lacks a focus. Although there are numerous conferences, symposia and journals emerging in this space there remains much work to be done to bring the disparate communities together and to corral their efforts to persuade funding bodies and governments of the importance of this field.
The time is ripe for bold initiatives. At a national level, governments need to work with academic researchers and industry to create mission-oriented research centres to accelerate the development of neuromorphic technologies. This has worked well in areas such as quantum technologies and nanotechnology—the US National Nanotechnology Initiative demonstrates this very well10, and provides focus and stimulus. Such centres may be physical or virtual but must bring together the best researchers across diverse fields. Their approach must be different from that of conventional electronic technologies in which every level of abstraction (materials, devices, circuits, systems, algorithms and applications) belongs to a different community. We need holistic and concurrent design across the whole stack. It is not enough for circuit designers to consult computational neuroscientists before designing systems; engineers and neuroscientists must work together throughout the process to ensure as full an integration of bio-inspired principles into hardware as possible. Interdisciplinary co-creation must be at the heart of our approach. Research centres must house a broad constituency of researchers.
Alongside the required physical and financial infrastructure, we need a trained workforce. Electronic engineers are rarely exposed to ideas from neuroscience, and vice versa. Circuit designers and physicists may have a passing knowledge of neurons and synapses but are unlikely to be familiar with cutting-edge computational neuroscience. There is a strong case to set up Masters courses and doctoral training programmes to develop neuromorphic engineers. UK research councils sponsor Centres for Doctoral Training (CDTs), which are focused programmes supporting areas with an identified need for trained researchers. CDTs can be single- or multi-institution; there are substantial benefits to institutions collaborating on these programmes by creating complementary teams across institutional boundaries. Programmes generally work closely with industry and build cohorts of highly skilled researchers in ways that more traditional doctoral programmes often do not. There is a good case to be made to develop something similar, to stimulate interaction between nascent neuromorphic engineering communities and provide the next generation of researchers and research leaders. Pioneering examples include the Groningen Cognitive Systems and Materials research programme, which aims to train tens of doctoral students specifically in materials for cognitive (AI) systems11,the Masters programme in neuroengineering at the Technical University of Munich12; ETH Zurich courses on analogue circuit design for neuromorphic engineering13; large-scale neural modelling at Stanford University14; and development of visual neuromorphic systems at the Instituto de Microelectrónica de Sevilla15. There is scope to do much more.
Similar approaches could work at the trans-national level. As always in research, collaboration is most successful when it is the best working with the best, irrespective of borders. In such an interdisciplinary endeavour as neuromorphic computing this is critical, so international research networks and projects undoubtedly have a part to play. Early examples include the European Neurotech consortium16, focusing on neuromorphic computing technologies, as well as the Chua Memristor Centre at the University of Dresden17, which brings together many of the leading memristor researchers across materials, devices and algorithms. Again, much more can and must be done.
How can this be made attractive to governments? Government commitment to more energy-efficient bio-inspired computing can be part of a broader large-scale decarbonization push. This will not only address climate change but also will accelerate the emergence of new, low-carbon, industries around big data, IoT, healthcare analytics, modelling for drug and vaccine discovery, and robotics, amongst others. If existing industries rely on ever more large-scale conventional digital data analysis, they increase their energy cost while offering sub-optimal performance. We can instead create a virtuous circle in which we greatly reduce the carbon footprint of the knowledge technologies that will drive the next generation of disruptive industries and, in doing so, seed a host of new neuromorphic industries.
If this sounds a tall order, consider quantum technologies. In the UK the government has so far committed around £1 billion to a range of quantum initiatives, largely under the umbrella of the National Quantum Technologies Programme. A series of research hubs, bringing together industry and academia, translate quantum science into technologies targeted at sensors and metrology, imaging, communications and computing. A separate National Quantum Computing Centre builds on the work of the hubs and other researchers to deliver demonstrator hardware and software to develop a general-purpose quantum computer. China has established a multi-billion (US) dollar Chinese National Laboratory for Quantum Information Sciences, and the USA in 2018 commissioned a National Strategic Overview for Quantum Information Science18, which resulted in a five-year US$1.2 billion investment, on top of supporting a range of national quantum research centres19. Thanks to this research work there has been a global rush to start up quantum technology companies. One analysis found that in 2017 and 2018 funding for private companies reached $450 million20. No such joined-up support exists for neuromorphic computing, despite the technology being more established than quantum, and despite its potential to disrupt existing AI technologies on a much shorter time horizon. Of the three strands of future computing in our vision, neuromorphic is woefully under-invested.
Finally, some words about what bearing the COVID-19 pandemic might have on our arguments. There is a growing consensus that the crisis has accelerated many developments already under way: for example, the move to more homeworking. Although reducing commuting and travel has direct benefits—some estimates put the reduction in global CO2 as a result of the crisis at up to 17%21—new ways of working have a cost. To what extent will carbon savings from reduced travel be offset by increased data centre emissions? If anything, the COVID pandemic further emphasizes the need to develop low-carbon computing technologies such as neuromorphic systems.
Our message about how to realize the potential of neuromorphic systems is clear: provide targeted support for collaborative research through the establishment of research centres of excellence; provide agile funding mechanisms to enable rapid progress; provide mechanisms for close collaboration with industry to bring in commercial funding and generate new spin-outs and start-ups, similar to schemes already in place for quantum technology; develop training programmes for the next generation of neuromorphic researchers and entrepreneurs; and do all of this quickly and at scale.
Neuromorphic computing has the potential to transform our approach to AI. Thanks to the conjunction of new technologies and a massive, growing demand for efficient AI we have a timely opportunity. Bold thinking is needed, and bold initiatives to support this thinking. Will we seize the opportunity?