Intel’s neurochips could one day end up in PCs or a cloud service

The brain-like chip technology could aid with low-power AI tasks like speech recognition


You may have heard before about Intel's Loihi neuromorphic chips that mimic the way brains work, but what hasn't been clear yet is how the chipmaker will make money from the experimental silicon.

In a recent roundtable with journalists, Intel Labs lead Rich Uhlig offered two possibilities: integrating Loihi in a CPU for PCs to perform energy-efficient AI tasks and potentially offering the its neuromorphic chips as a cloud service, although Uhlig was clear he wasn't firming actual product plans, just projecting what could theoretically happen in the future.

"Right now with Loihi, we're at that point where we think we're onto something, but we don't actually have product plans yet. We're sort of earlier on in that work stream," he said last month.

Intel's Loihi neurochips have been in development for the past several years under the purview of Intel Labs, which is insulated from commercial pressures inside the semiconductor giant so that it can conduct advanced research of "everything from the circuits up," as Uhlig put it. This includes everything from microarchitecture and system architecture to system software and compilers.

While Intel Labs' budget is protected, the goal is to create technologies that will someday turn into products. The group has churned out plenty of commercial technologies in the past, including Thunderbolt, hardware-assisted virtualization, and Intel Software Guard Extensions.

Since Intel first revealed its Loihi chips in 2017, the company has pitched them as a new way to handle AI tasks faster using far less energy. This is made possible by the fact that the neurochips' pin-like structures, which are meant to replicate neurons in a brain, only consume electricity when they contribute data. That's in contrast to traditional processors, where energy is used regardless of whether all the circuits are processing information.

This breakthrough in energy-efficient computing allows Loihi, for instance, to handle image similarity searches 24 times faster than a CPU using only 30 times less energy. The neurochip can also solve complex optimization problems like Sudoku puzzles or train scheduling 44 times faster than a CPU using 2,800 times less energy, another example offered by Intel.

Recently, researchers at the United States' Sandia National Laboratories found that Loihi is a good fit for high-performance computing applications, too, like tracking financial market movements, disease spread within a population, and data flows within social networks.

These use cases represent things that could eventually generate revenue for Intel, but when Uhlig was asked how Loihi could materialize as an actual product, he pointed to the PC and cloud examples.

For the PC example, Uhlig said Intel could feasibly use its neuromorphic chip technology as a building block for silicon in a client system — we're thinking CPUs in this case. The purpose would be to perform common AI tasks like speech recognition or gesture recognition.

While the use of neuromorphic technology in a CPU would be new, there is precedent for Intel integrating specialized accelerator technologies in its processors. One of the more prominent examples for PCs is the Gaussian Neural Accelerator that was introduced in Intel's 10th-generation Core processors to handle low-power tasks like real-time translation.

As for offering neurochips as a cloud service, Uhlig said Intel Labs is now in a better place organizationally where it can stand up such products in the future. That's because Intel Labs was moved last year under the Software and Advanced Technology Group, which is led by Intel CTO Greg Lavender.

The new organizational structure is a benefit for Intel, Uhlig said, "because we're able to take our innovation, particularly things that are software-oriented, and scale them through Greg's organization," which has the charter of identifying new software revenue opportunities, as The Register has recently reported.

When it comes to how Intel's Loihi's development could benefit the company in other ways, Uhlig said the chipmaker could take the design methodology used for neurochips, called asynchronous design, and apply it to the way other chips are made.

"What we can do is take that asynchronous design methodology and deploy it into other parts of Intel products and get additional efficiency," he said. ®


Other stories you might like

  • Intel withholds Ohio fab ceremony over US chip subsidies inaction
    $20b factory construction start date unchanged – but the x86 giant is not happy

    Intel has found a new way to voice its displeasure over Congress' inability to pass $52 billion in subsidies to expand US semiconductor manufacturing: withholding a planned groundbreaking ceremony for its $20 billion fab mega-site in Ohio that stands to benefit from the federal funding.

    The Wall Street Journal reported that Intel was tentatively scheduled to hold a groundbreaking ceremony for the Ohio manufacturing site with state and federal bigwigs on July 22. But, in an email seen by the newspaper, the x86 giant told officials Wednesday it was indefinitely delaying the festivities "due in part to uncertainty around" the stalled Creating Helpful Incentives to Produce Semiconductors (CHIPS) for America Act.

    That proposed law authorizes the aforementioned subsidies for Intel and others, and so its delay is holding back funding for the chipmakers.

    Continue reading
  • Microsoft promises to tighten access to AI it now deems too risky for some devs
    Deep-fake voices, face recognition, emotion, age and gender prediction ... A toolbox of theoretical tech tyranny

    Microsoft has pledged to clamp down on access to AI tools designed to predict emotions, gender, and age from images, and will restrict the usage of its facial recognition and generative audio models in Azure.

    The Windows giant made the promise on Tuesday while also sharing its so-called Responsible AI Standard, a document [PDF] in which the US corporation vowed to minimize any harm inflicted by its machine-learning software. This pledge included assurances that the biz will assess the impact of its technologies, document models' data and capabilities, and enforce stricter use guidelines.

    This is needed because – and let's just check the notes here – there are apparently not enough laws yet regulating machine-learning technology use. Thus, in the absence of this legislation, Microsoft will just have to force itself to do the right thing.

    Continue reading
  • Cerebras sets record for 'largest AI model' on a single chip
    Plus: Yandex releases 100-billion-parameter language model for free, and more

    In brief US hardware startup Cerebras claims to have trained the largest AI model on a single device powered by the world's largest Wafer Scale Engine 2 chip the size of a plate.

    "Using the Cerebras Software Platform (CSoft), our customers can easily train state-of-the-art GPT language models (such as GPT-3 and GPT-J) with up to 20 billion parameters on a single CS-2 system," the company claimed this week. "Running on a single CS-2, these models take minutes to set up and users can quickly move between models with just a few keystrokes."

    The CS-2 packs a whopping 850,000 cores, and has 40GB of on-chip memory capable of reaching 20 PB/sec memory bandwidth. The specs on other types of AI accelerators and GPUs pale in comparison, meaning machine learning engineers have to train huge AI models with billions of parameters across more servers.

    Continue reading
  • Intel demands $625m in interest from Europe on overturned antitrust fine
    Chip giant still salty

    Having successfully appealed Europe's €1.06bn ($1.2bn) antitrust fine, Intel now wants €593m ($623.5m) in interest charges.

    In January, after years of contesting the fine, the x86 chip giant finally overturned the penalty, and was told it didn't have to pay up after all. The US tech titan isn't stopping there, however, and now says it is effectively seeking damages for being screwed around by Brussels.

    According to official documents [PDF] published on Monday, Intel has gone to the EU General Court for “payment of compensation and consequential interest for the damage sustained because of the European Commissions refusal to pay Intel default interest."

    Continue reading

Biting the hand that feeds IT © 1998–2022