There is vast potential in this field. A recent paper published in Nature suggested that neuromorphic computing is ready to transition into real-world environments, highlighting the efficiency of the technology in areas like Internet of Things and edge processing. Given the power-hungry nature of GPUs used in training AI models, neuromorphic systems could help alleviate some of these energy demands.
SpiNNcloud envisions a range of applications for SpiNNaker2, from small multilayer perceptrons for molecule identification in drug discovery to QUBO-based optimization solving various complex mathematical simulations. The system is designed to support generative AI algorithms through dynamic sparsity, accommodating recent developments in machine learning toward extreme dynamic sparsity to address energy efficiency challenges in AI computing.
Lately, the focus seems to have shifted from the technology itself towards finding ways to push it beyond its current niche status into mainstream adoption. Proof-of-concepts are underway, systems are being scaled up, and the quest is on to identify the so-called “killer app,” the breakthrough application that will drive widespread adoption of this technology.
- The development of this technology has been closely followed by The Next Platform, exploring various aspects such as potential operating systems, military applications, the need for strategic partnerships, and software development to make practical use of neuromorphic computing. Just like with other revolutionary computing concepts
- think quantum computing
- progress takes time.
The highly parallel architecture of SpiNNaker2 comprises 24 boards, each housing 48 SpiNNaker2 chips interconnected in toroidal topologies. Each chip features 152 Arm-based low-power processing elements networked on a chip-on-chip architecture. Unlike traditional neuromorphic systems, SpiNNaker2 does not rely on spiking neurons, allowing for fine-grained control over its 175K cores and enabling scalability in neural symbolic models.
The deployment of SpiNNaker2 at Sandia National Laboratories is a significant milestone, with the system simulating around 175 million neurons and ranking among the top five computing platforms designed to mimic the human brain’s functionality.
Companies like Intel, IBM, and Google have been at the forefront of developing neuromorphic computing alongside a growing number of smaller startups in recent decades, aiming to replicate the structure and function of the human brain.
Sandia Labs has been actively exploring neuromorphic computing, incorporating technologies like Intel’s Loihi 2 processor in their AI research. SpiNNaker2 joins their arsenal as part of ongoing efforts to leverage energy-efficient architectures for AI applications, aiming for reduced power consumption compared to conventional GPU-based systems.
SpiNNcloud, a company founded four years ago from Dresden University of Technology, claims that its SpiNNaker2 chip architecture offers ultra energy-efficient infrastructure for next-generation AI inference, boasting efficiency levels significantly higher than existing GPU-based systems. Their upcoming iteration, SpiNNext, is projected to be even more efficient.
