Intel details Loihi AI and 49-qubit quantum computing chips


Quantum computing may still be a long way from commercial viability, but Intel is making great strides in scaling up the technology. During a CES 2018 keynote, the company unveiled its 49-qubit superconducting quantum test chip called “Tangle Lake”. During the same keynote, Intel CEO Brian Krzanich also noted the promise of neuromorphic computing – a computing paradigm inspired by the human brain that the company is developing under the Lohi code-name.

The announcement of Tangle Lake comes just two months after the company delivered its cutting edge 17-qubit test chip to the to Dutch research centre, QuTech. The move to a working 49-qubit design moves researchers closer to perfecting a commercially useful chip, allowing them to improve error correction techniques, simulate more intensive computational problems, and also reducing radio frequency interference. However, Intel Labs managing director Mike Mayberry states that a commercially relevant chip will likely require one million or more qubits, so there’s still clearly a long way to go.

To potentially expedite development of larger chips, Intel has begun investigating spin qubits, addition to superconducting qubits. Spin qubits are much smaller than the superconducting type and, as they resemble a single electron transistor, it’s possible that they could be fabricated on a comparable process to conventional transistors. Intel says that it already has a spin qubit fabrication flow on its 300mm process technology.

In addition to quantum computing, artificial intelligence will no doubt continue to be a buzzword throughout 2018. Intel, along with a number of other silicon designers, are continuing to work on chips specifically designed for efficient AI workloads. The company’s new Loihi research chip is rather interesting in that it’s based on a neuromorphic circuit designed to mimic the brain’s basic operations.

Neuromorphic computing does away with the traditional CPU and memory chip design, and instead adopts a networked design where data flows in a highly parallel fashion between digital neurons” Rather than constant clocked data, neuromorphic chips use pulses and spikes between synapses, with different parts of the chip taking on different learning functions. Intel claims that Loihi is able to adjust the network as data is passed through, allowing it to “learn” as time goes by. This will sound familiar to those who have read up on Machine Learning and Neural Networks.

Intel showcased its first Loihi chip at CES, which is built on the company’s 14nm process and packs in over 130,000 digital neurons and synapses. The chip is scheduled to head out to universities and AI research centers sometime in the first half of 2018.

With quantum computing and AI research firmly under its belt, Intel is clearly prepared for the next generation of big data, machine learning, and AI.

Leave a comment