Human communication

Intel’s Loihi 2 steps up efforts to make neuromorphic chips like human brains

Intel’s Loihi 2 neuromorphic chip measures 30 square millimeters.


Intel on Thursday unveiled its Loihi 2 chip, the second generation in a family of processors that combines conventional electronics with the architecture of the human brain in an attempt to inject new advancements into the computer industry. In addition to that, the chip also helps Intel advance its own manufacturing technology.

Loihi 2, an example of a technology called neuromorphic computing, is about 10 times faster than its predecessor, according to Intel. The improvement in speed is the result of an eight-fold increase in the number of digital neurons, a chip equivalent to human brain cells that mimic how the brain handles information. The chip may also be better programmed to help researchers tackle more computing tasks.

The chip is built with a pre-production version of the Intel 4 manufacturing process, too, an advanced method that Intel plans to use to build consumer Intel chips coming in 2023. The Intel 4 process can burn electronics more densely on a chip, a crucial advantage for Intel’s need to wrap a million digital neurons on a chip measuring 30 square millimeters.

Loihi chips are particularly good at quickly detecting sensory input such as gestures, sounds and even smells, says Mike Davies, head of the Intel Labs group that developed Loihi. Some experiments focused on artificial skin which could give robots a better sense of touch. “We can detect a slip if a robot hand picks up a cup,” Davies said.

Neuromorphic computing differs from artificial intelligence, a revolutionary computer technology based more freely on the way brains learn and react, as it focuses more on the physical characteristics of human gray matter.

It differs profoundly from conventional chips. For example, Loihi 2 stores data in small amounts spread over its mesh of neurons, not in a large traditional computer memory bank, and it does not have a central clock to synchronize the computational steps on the chip.

You will not see Loihi 2 on your phone or laptop. Instead, it is aimed at researchers at car manufacturers, national laboratories, and universities. Deutsche Bahn’s German rail network is testing how well it can optimize train schedules. The processor is designed for tasks like processing sound or detecting hand gestures, but with significantly lower power consumption, Davies said.

Neuromorphic calculus

Low energy consumption is also a characteristic of biological gray matter. The human brain is made up of around 80 billion cells called neurons, connected to elaborate electrical signaling networks. When enough input signals reach an individual neuron, it sends its own signal to the other neurons. The topology of connections and signal flow allows us to do everything from recognizing Abraham Lincoln to riding a bicycle. Learning is the process of making and strengthening these connections.

Intel isn’t the only one pursuing the idea. The Human Brain Project in Europe includes neuromorphic informatics in its work. The way blood circulates in the brain inspired IBM to Power and freshness chips with liquids in a flow coil. Samsung used IBM’s TrueNorth neuromorphic chip to recreate a vision.

Intel’s chip is made up of a million digital neurons that can be connected in several ways, a digital clean slate. To make it work requires configuring the right connections between neurons. The actual processing occurs when the input data reaches the chip, triggering a spike in activity that passes through the interconnected neurons and ultimately produces an output. Each neuron is connected to 100 others on average, although some can grow as high as 10,000.

This fluid design means the chip requires very little power when idle and can process data very quickly on demand, Davies said.

Programming neuromorphic chips is a big challenge, Davies said. To try to make the process easier for researchers, Intel has also released an open source software framework called Lava.

Fewer but smarter neurons

A million neurons in a chip is a far cry from billions in a human brain, but Intel is actually trying to make every neuron smarter than a biological brain cell. For example, in biological brains, electrical signals are either completely turned on or completely turned off. In Loihi chips, Intel can assign a different strength to each signal, increasing the sophistication of the processing, Davies said.

The chip can also be connected to others for a larger scale. An improvement over the first Loihi is better networking which shortens the communication pathways that connect neurons.

“The brain achieves precision and reliability through enormous redundancy,” Davies said. “The hope is indeed that we can solve some of the same problems in a more economical way.”

Intel 4 manufacturing

The Intel Process 4 is a major milestone for Intel, its first step towards a chip-making technology called Extreme Ultraviolet, or EUV. The chip circuits are etched onto silicon wafers using light patterns, and the UVU allows finer patterns for smaller structures, crucial for miniaturization and other enhancements. The problems of the past few years have caused Intel to lose its manufacturing leadership to Samsung and Taiwan Semiconductor Manufacturing Co. (TSMC), and Intel 4 is a key part of Intel’s efforts to reclaim that leadership. by 2025.

Consumer manufacturing with Intel 4 won’t begin until the second half of 2022, with the resulting chips arriving in 2023, Intel said. But Loihi 2 gives the company a chance to launch the technology while it is still in the pre-production stage.

“We have small amounts in the lab now,” Davies said.

Intel has made thousands of first-generation Loihi chips, Intel said, and Loihi 2 will likely mean similarly small production. Manufacturing at the full scale of Intel’s operations, with millions of processors, presents other challenges.

Leave a Reply

Your email address will not be published.