A few days ago Intel made one of those announcements that usually go unnoticed by ordinary mortals, but which for scientists, researchers and technology experts is a milestone to be highlighted. It unveiled in mid-July a neuromorphic system of eight million neurons made up of 64 Loihi research chips, called Pohoiki Beach, and whose design is inspired by the human brain. This chip, or set of processors, resembles the principles of biological brains to computer architectures. Loihi enables users to process information up to 1,000 times faster and 10,000 times more efficiently than central processing units (CPUs) used in specialized applications, including sparse coding, graphics search, and constraint compliance issues.
Thanks to the introduction of Pohoiki Beach, researchers will be able to efficiently scale their new algorithms inspired by neural networks, such as sparse coding, simultaneous localization and modeling, and route planning, capable of learning and adapting based on the input data. . Intel’s new chip system represents an important milestone in neuromorphic research work and lays the foundation for the Intel Labs team’s goal of increasing its architecture to 100 million neurons by the end of this year.
Intel says that “continuing the capacity and performance improvements established by Moore’s Law will require more than a continuous process of increasing nodes. As new complex computing workloads become the norm, so does the need for specialized architectures designed for specific applications. “
The Pohoiki Beach neuromorphic system demonstrates the benefits of a specialized architecture for emerging applications, including some of the toughest computing problems related to the Internet of Things (IoT) and autonomous devices. By employing these types of specialized systems, rather than general-purpose computing technologies, gains of several orders of magnitude in speed and efficiency can be expected in a wide range of real-world applications, such as autonomous vehicles, smart homes, and cybersecurity. .
Loihi systems applications
Researchers are using Loihi systems to solve challenges at the forefront of neuromorphic engineering, such as providing adaptive capabilities to the AMPRO prosthetic leg, camera object tracking based on emergent events, automation of a table football through perception and neuromorphic control, learn to control an inverted linear pendulum and infer tactile data from the electronic skin of an iCub robot.
“ Thanks to the Loihi chip we have been able to demonstrate 109 times lower consumption in a real-time deep learning standard, compared to a graphics processing unit (GPU), and up to 5 times lower compared to specialized hardware for IoT inference”, explains Chris Eliasmith, CEO of Applied Brain Research and Professor at the University of Waterloo. “What’s more, as we scale up the network fifty times, Loihi maintains performance results in real time requiring only 30% more power consumption, while IoT hardware consumes 500% more and stops operating in real time”.
The future
In 2017, Intel introduced Loihi, its first neuromorphic research chip, which introduced a major breakthrough in neuromorphic hardware development. In March 2018, the company established the Intel Neuromorphic Research Community (INRC) to drive the development of neuromorphic algorithms, software, and applications. Through INRC Intel provides access to Loihi systems in the cloud, as well as Kapoho Bay, a USB form factor Loihi-based system that has accelerated research on real-world applications of neuromorphic technologies.
Later this year Intel will introduce an even larger Loihi system called Pohoiki Springs, which will build on the Pohoiki Beach architecture to deliver unprecedented performance and efficiency for large-scale neuromorphic workloads. Intel engineers hope that the results of these research systems will make it possible to quantify the gains generated by the new methods of neuromorphic computing and help to clarify the most suitable application areas for this technology. This research paves the way for an eventual commercialization of neuromorphic technology and, perhaps, for non-biological intelligence to be closer to human.
