New Brain-inspired Supercomputers for the Next Generation of AI

The brain-inspired “mesh” computing concept

The concept of using a large-scale “mesh” composed of constrained low-power nodes was originally developed by the University of Manchester in the UK. This was then further enhanced by TU Dresden, which incorporated functional principles of how the human brain works into modern silicon. The mesh computing concept revolves around the idea that the human brain is composed of a large number of small computing nodes that operate in an event-based nature and whose connections adapt to learn new concepts, similar to the human brain. These functional principles are highly relevant in addressing the challenges faced by modern artificial intelligence (AI) models.

The “Human Brain” project

During the inception of the “Human Brain Project” in 2013, these ideas were introduced by Professor Steve Furber from the University of Manchester, who was one of the original creators of the first Arm processor, the ARM1. They were further expanded and enhanced by Dr. Christian Mayr’s research team at TU Dresden. The aim of their collaboration was to leverage the brain-inspired mesh concept on a large scale to achieve low power AI across various scenarios. This collaboration eventually led to a commercial spinout, with the startup company named SpiNNcloud Systems being established in 2021.

The SpiNNaker products

The first product based on this research philosophy was the SpiNNaker1 (Spiking Neural Network Architecture) microchip, which was led by Professor Steve Furber. This Arm-based microchip was used to create a supercomputer comprising one million cores that supports real-time modeling of brain simulations. This was a pioneer in the neuromorphic field and instrumental in advancing the study of the brain within the field of neuroscience, sparking interest from national labs and R&D centers.

SpiNNcloud then took the mesh computing concept from both universities to a whole new level with the next-generation SpiNNaker2, which delivers algorithms that are more efficient than conventional AI. The hybrid AI processor is used to build a ten million core supercomputer that adopts a large mesh of interconnected processors powered by Arm Cortex-M4 cores in a lightweight network on chip. Also integrated into this mesh are accelerators designed to execute event-based deep neural networks (DNNs), which is a pioneering approach to combine neuromorphic principles with conventional DNNs to achieve significant savings.

SpiNNaker2 delivers large scale AI in real-time with an extremely high level of energy efficiency for a variety of use cases, including complex production optimization and logistics, pharmaceutical discovery and innovation, novel large language models (LLMs) and robotics, as well as smart farming and cognitive city applications. There is already strong interest from companies and research institutions in the product and wider brain-inspired architecture as a way to speed up computing and save costs.

This is reflected in the impressive metrics of the SpiNNaker2 microchip, which already shows processing that is 10 times more energy efficient compared to the high-end, state-of-the-art GPUs traditionally used for high-performance computing (HPC). Additionally, it enables brain-like connectivity that is impossible to achieve with these GPUs.

Héctor González, co-Founder and co-CEO at SpiNNcloud, believes the company has a clear strategic role in helping to reduce the energy footprint when executing various AI workloads. He says: “Key strategic partners and the entire Arm technology ecosystem are transitioning to a future where AI models are deployed in the most energy efficient way possible. This is the value of the SpiNNaker2 and the mesh concept for large-scale AI.”

Low power leadership, flexible licensing models

Arm’s leadership in low power, power efficient technology is helping to deliver solutions for large scale efficient AI that scales. While Cortex-M is not a typical product line for large scale AI, it was the ideal IP processor for the low power capabilities required for the mesh concept that is utilized in the SpiNNaker2. This is one of the key aspects differentiating SpiNNcloud’s approach to conventional HPC.

González says: “We have this large mesh of processing elements where we aim to reduce the overall footprint in terms of area and power. The Cortex-M line exactly meets those requirements. Arm’s technology is also proven across a variety of different markets, many of which we are looking to target through the SpiNNaker2.”

Alongside Arm’s technology capabilities, González is especially complimentary about what he describes as a “unique” ecosystem. He says: “One of the key benefits to working with Arm is the ecosystem. It is incredibly unique with a wide range of partners, from top-tier technology companies to millions of developers.”

Arm’s technology capabilities and unique ecosystem combined with the Arm Flexible Access for Startups program, which delivers early, zero-cost access to Arm IP during prototype development and wider technical support, means early-stage startups can move quickly with their innovative processor designs. SpiNNcloud’s relationship with Arm began with Arm Academic Access during the “Human Brain Project”, but as the company was spun out from the research teams it transitioned to Arm Flexible Access for Startups.

González says: “The early, zero cost access to the Arm IP during prototype development through Arm Flexible Access for Startups was very helpful for SpiNNcloud. Normally startups would have to pay up front, so this helped the company manage our cash flow during the early stages of product development. The program, which includes the vast Arm ecosystem support, also provided us with a vehicle to commercialize our products straight away.”

Bringing innovative AI tech to life

The story of SpiNNcloud highlights the benefits that Arm brings to early-stage startups through Arm Flexible Access for Startups. The program is lifting startups like SpiNNcloud to a technology level that is comparable with other leading technology companies, helping to make the company’s innovative vision of brain-inspired AI on silicon a reality.

Learn more about Arm Flexible Access for Startups here.