The constant increase in data traffic (22% more last year compared to 2022, according to DE-CIX) and the new computational demands of artificial intelligence push conventional systems to the limit. New formulas are needed and quantum computing is not yet an alternative. The electronics company Intel is one of the most advanced in the development of neuromorphic systems, a conjunction of biology and technology that seeks to imitate the way in which human beings process information. Along with her, in this race for more effective and efficient processing, they run IBM, Qualcomm and research centers such as California Institute of Technology (Caltech)where this concept was born by Carver Mead, the MIT (Massachusetts Institute of Technology)he Max Planck Institute for Neurobiology in Germany and Stanford University.
Intel this month announced the world’s largest neuromorphic system: Hala Pointwith 1,150 million technological neurons and 1,152 processors (chips) Loihi 2 that consume a maximum of 2,600 watts and with a processing capacity equivalent to that of an owl’s brain. A study published in IEEE Xplore It attributes greater efficiency and performance than systems based on central processing units (CPU) and graphics units (GPU), conventional computing engines.
Mike Davis, born in Dallas and who turns 48 in July, is director of neuromorphic computing at Intel Labs and most responsible for the latest advances on which the immediate future of computing rests.
Ask. What is a neuromorphic system?
Answer. It is a computing design that is inspired by modern understanding of how brains work and that means surpassing seven or eight decades of conventional architecture. From a basic perspective, we are trying to understand the principles of modern neuroscience to apply it to chips and systems in order to create something that operates and processes information more like the way a brain works.
Q. How does it work?
R. If you open the system, the chips, you see very striking differences in the sense that there is no memory; All computing, processing and memory elements are integrated with each other. Our Hala Point system, for example, is a three-dimensional network of chips similar to a brain and everything communicates with everything, just as one neuron communicates through the brain with another set of connected neurons. In a traditional system, you have memory next to a processor and the processor continually reads out of memory.
Hala Point is a three-dimensional network of chips similar to a brain and everything communicates with everything, like a neuron communicates through the brain with another set of connected neurons.
Q. Is this model necessary because we are reaching the limit of conventional computing?
R. Much progress is being made in artificial intelligence and deep learning. It’s very exciting, but it’s hard to see how these research trends will continue when the increase in computing requirements for these AI models is growing at exponential rates—that is, much faster than manufacturing advances. We are reaching the limits of what this basic computing architecture can provide. Also, if you just look at the power efficiency of these traditional AI chips and systems compared to the brain, there are many orders of magnitude difference. It is not so much that traditional computer architectures are not capable of providing great gains in computing and artificial intelligence, but rather that we are looking for greater functionality, by having computers that operate like the brain, and do so in a very efficient way.
Q. Is energy efficiency the main advantage?
R. It is one of the main ones. There is a big difference in the efficiency of the brain and that of traditional computing. But brain-inspired neuromorphic architectures can also provide performance benefits. We think of GPUs as incredibly high performance devices, but in fact only if you have a very large size and a lot of data to process available on disk or right next to the processor to be read. But if the data comes from sensors, cameras or videos in real time, then the efficiency and power of traditional architectures is much lower. That’s where neuromorphic architectures can really provide a huge increase in speed as well as efficiency.
Q. Does artificial intelligence need a neuromorphic system to grow?
R. We think so. But we are at a research level. It is unclear today how to implement this commercially. There are many problems still to be solved regarding the software [programación], the algorithms. Many conventional approaches do not run natively on hardware [equipos] neuromorphic because it is a different programming approach. We believe this is the right path to achieve the gains we need in power efficiency and performance for these types of workloads, but it remains an open question.
Q. Will we see a neuromorphic chip in a computer or mobile phone?
R. I think so, it’s a matter of time. It won’t be next year, but the technology will mature and be implemented in edge computing [procesamiento de datos cerca de su origen para ganar velocidad y eficiencia], mobile phones, autonomous vehicles, drones or on the laptop. Our Hala Point, designed for a data center, is a box the size of a large microwave. But, if we look at nature, we see that there are brains of all sizes. The insect ones are very impressive, even on a small scale. And then there is, of course, the human brain. We are pursuing both directions in the investigation. We believe that commercialization will begin in the edge computingbut there is a need to continue pushing and doing research on a large scale.
In data centers we could see these systems in five years
Q. When will they be?
R. It is difficult to predict because there are still open questions in the research. In data centers we could see these systems in five years. We also see a future in anything that needs to be battery powered, as the energy savings that a neuromorphic system can offer is extremely important. There are also somewhat less obvious applications, such as wireless base stations for telephone infrastructure. We are working with Ericsson to optimize communication channels.
Q. Is computing with neuromorphic systems complementary to quantum computing?
R. I think they are complementary in some ways, although they are very different. Quantum computing is seeking innovation in physical device manufacturing and trying to scale. What it offers is very novel and impressive, but it is not clear what the programming model of quantum will be once it can be scaled and what kind of loads it will support. The neuromorphic computing available today is very good for AI type of workloads. But there is an intersection in the quantum and neuromorphic application space and that’s where it’s interesting to think about solving difficult optimization problems and allowing people to experiment, prototype and learn to program these types of systems.
Implanting neuromorphic chips in the brain is a very natural application of these systems because, being an architecture that behaves like neurons, it would naturally speak the language of our brain.
Q. Could we see neuromorphic systems installed in our brain?
R. There are some researchers interested in neuroprosthetics, in the application of neuromorphic computing, which would mean trying to repair problems or pathologies in the brain where there has been some loss of function and returning control over your body. The research is in an early stage, but I think that, in the long term, it is a very natural application of neuromorphic computing because, being an architecture that behaves like neurons, it would naturally speak the language of our brain.
Q. The available systems, what age brain are they equivalent to?
R. In terms of the number of neurons, it is similar to the owl brain. But if you look at the area of the brain where much of the higher-order intelligence occurs, it would be equivalent to a capuchin monkey. Many of us, in this field of research, have the human brain in mind as a kind of vision for the scale of system we would like to build. But we don’t try to get there too quickly. We have to build it well and we need to know how to make it useful. That is why this system remains a tool for research, so that we can continue experimenting
Q. In what specific cases are these systems most effective?
R. Finding the best path on a map we see speedups of up to 50 times compared to the best conventional solvers. In terms of energy, levels 1,000 times more efficient are being reached.
Q. Could Europe take advantage of this new line to regain positions in the race for chips, on which it is dependent on other continents?
R. If we look to the future, there is a lot of innovation we will need in the long term to achieve the size and efficiency of nature, which remains incredibly impressive. There is still a long way to go and to get there we need innovation in manufacturing. There needs to be new devices and new memory technology to assimilate them into the brain. Now there is no geographic region with an advantage in this domain, so it is an opportunity. High technology always implies innovation and nothing remains static. There is a need for new advances and it is unpredictable where they may come from.
You can follow EL PAÍS Technology in Facebook and x or sign up here to receive our weekly newsletter.
