Source-linked AI summary
Brain-inspired computing: We need a master plan
Adnan Mehonic, Anthony J Kenyon
TL;DR
Current digital computing faces escalating energy demands from AI and other data-intensive applications, while the von Neumann separation of storage and processing contributes to this burden. The paper argues for brain-inspired systems that exploit biological principles and new devices, supported by coordinated interdisciplinary investment. It concludes that neuromorphic computing offers timely potential for more efficient AI and motivates bold initiatives to develop it.
Problem
Escalating computing demand, AI training costs, and environmental impacts expose limits in relying on digital von Neumann systems alone.
Method
The paper synthesizes biological inspiration, neuromorphic technologies, and research-policy needs into a coordinated plan for developing brain-inspired computing.
Results
Neuromorphic systems are presented as promising complementary platforms for energy-efficient processing of unstructured, noisy, and uncertain data.
Takeaways & Limitations
Realising neuromorphic computing’s potential requires rapid, large-scale collaboration, research centres, agile funding, industry links, and training.
Abstract
from arXiv · showhide
New computing technologies inspired by the brain promise fundamentally different ways to process information with extreme energy efficiency and the ability to handle the avalanche of unstructured and noisy data that we are generating at an ever-increasing rate. To realise this promise requires a brave and coordinated plan to bring together disparate research communities and to provide them with the funding, focus and support needed. We have done this in the past with digital technologies; we are in the process of doing it with quantum technologies; can we now do it for brain-inspired computing?
The problem
Modern digital computing is becoming environmentally unsustainable as computing demand and AI training costs grow faster than efficiency improvements. Brain-inspired approaches are proposed because they can process information differently from energy-intensive von Neumann systems.
- Computing power demand now doubles approximately every two months, compared with every 24 months before 2012.
- NVIDIA GPU performance improved 317-fold since 2012, exceeding what Moore’s law alone would predict.
- AI model-training costs have risen exponentially since 2011, making the trend unsustainable.
- Von Neumann systems consume energy moving data between separately stored data and processors.
- Brain-inspired platforms use alternatives including analogue processing, asynchronous communication, massive parallelism, and spiking representations.
- Neuromorphic devices such as memristors can implement memory, synaptic weights, cognitive processing, sensors, and neuron-like functions while using very little energy.
Biological inspiration
Biological systems motivate neuromorphic computing because the brain combines storage and processing while efficiently handling noisy, imprecise, and unstructured data. The field spans approaches from emulating neural function to applying selected biological principles in new hardware and systems.
- A human brain with approximately 10^10 neurons and 10^14 synapses uses about 20 W, whereas a similarly sized digital neural-network simulation consumes 7.9 MW.The comparison indicates a six-order-of-magnitude power gap.
- The brain processes noisy, imprecise, and unstructured data efficiently, whereas comparable tasks are costly for digital supercomputers.
- The field’s renewed interest reflects growing AI demand and new devices that can mimic biological neural capabilities.
- Neuromorphic chips incorporate features such as in-memory computing, spike-based processing, fine-grained parallelism, reduced precision, stochasticity, adaptability, and asynchronous communication.
- Neuromorphic technologies range from reverse-engineering brain structure and function to synthesizing new systems from selected biological principles.
Prospects
Neuromorphic systems are presented as complementary to digital and quantum computing, especially for unstructured, noisy, and uncertain data in edge and autonomous applications. Realising this potential requires interdisciplinary collaboration and coordinated support.
- Neuromorphic systems are intended to complement rather than replace digital computation.Digital systems remain suited to precise calculations, while neuromorphic systems target unstructured and uncertain data.
- Neuromorphic systems can process unstructured data, recognize images, classify noisy and uncertain datasets, and support learning and inference.
- A three-way synergy between digital, neuromorphic, and quantum systems is proposed.
- Neuromorphic computing requires contributions from physicists, chemists, engineers, computer scientists, biologists, and neuroscientists.
Seizing the opportunity
Realising neuromorphic computing’s potential requires coordinated, large-scale support spanning research communities, infrastructure, training, industry and international collaboration. The paper argues that this opportunity is timely because demand for efficient computing is rising while neuromorphic funding remains far below digital AI and quantum investment.
- Seizing the opportunity: Neuromorphic research lacks the focus, roadmap and investment scale of quantum computing and the semiconductor industry.Existing examples include IBM’s TrueNorth projects, Intel’s Loihi processor and the US Brain Initiative, but the sums remain below the paper’s assessment of need.
- Seizing the opportunity: Digital CMOS, quantum and neuromorphic systems could operate in parallel across different application domains, although neuromorphic computing has received the least attention.The paper presents this as a future hardware scenario rather than a replacement of existing systems.
- Seizing the opportunity: Mission-oriented research centres should unite academic and industry researchers across the full stack through holistic, concurrent design.The proposed centres may be physical or virtual, but interdisciplinary co-creation should integrate materials, devices, circuits, systems, algorithms and applications.
- Seizing the opportunity: Cross-border networks can strengthen neuromorphic research by connecting leading researchers across materials, devices and algorithms.The Chua Memristor Centre at the University of Dresden is cited as an early example bringing together memristor researchers across these areas.
- Seizing the opportunity: Government support for energy-efficient bio-inspired computing could contribute to decarbonisation and low-carbon industries in big data, IoT, healthcare analytics, drug discovery and robotics.The paper links this policy rationale to increased computing demand and the need to reduce energy requirements across emerging applications.
- Seizing the opportunity: The field needs agile funding, industry collaboration, spinouts, start-ups and training programmes delivered quickly and at scale.The paper also recommends research centres of excellence and mechanisms modelled on support schemes used for quantum technologies.