Tag Archives: tiny
#435541 This Giant AI Chip Is the Size of an ...
People say size doesn’t matter, but when it comes to AI the makers of the largest computer chip ever beg to differ. There are plenty of question marks about the gargantuan processor, but its unconventional design could herald an innovative new era in silicon design.
Computer chips specialized to run deep learning algorithms are a booming area of research as hardware limitations begin to slow progress, and both established players and startups are vying to build the successor to the GPU, the specialized graphics chip that has become the workhorse of the AI industry.
On Monday Californian startup Cerebras came out of stealth mode to unveil an AI-focused processor that turns conventional wisdom on its head. For decades chip makers have been focused on making their products ever-smaller, but the Wafer Scale Engine (WSE) is the size of an iPad and features 1.2 trillion transistors, 400,000 cores, and 18 gigabytes of on-chip memory.
The Cerebras Wafer-Scale Engine (WSE) is the largest chip ever built. It measures 46,225 square millimeters and includes 1.2 trillion transistors. Optimized for artificial intelligence compute, the WSE is shown here for comparison alongside the largest graphics processing unit. Image Credit: Used with permission from Cerebras Systems.
There is a method to the madness, though. Currently, getting enough cores to run really large-scale deep learning applications means connecting banks of GPUs together. But shuffling data between these chips is a major drain on speed and energy efficiency because the wires connecting them are relatively slow.
Building all 400,000 cores into the same chip should get round that bottleneck, but there are reasons it’s not been done before, and Cerebras has had to come up with some clever hacks to get around those obstacles.
Regular computer chips are manufactured using a process called photolithography to etch transistors onto the surface of a wafer of silicon. The wafers are inches across, so multiple chips are built onto them at once and then split up afterwards. But at 8.5 inches across, the WSE uses the entire wafer for a single chip.
The problem is that while for standard chip-making processes any imperfections in manufacturing will at most lead to a few processors out of several hundred having to be ditched, for Cerebras it would mean scrapping the entire wafer. To get around this the company built in redundant circuits so that even if there are a few defects, the chip can route around them.
The other big issue with a giant chip is the enormous amount of heat the processors can kick off—so the company has had to design a proprietary water-cooling system. That, along with the fact that no one makes connections and packaging for giant chips, means the WSE won’t be sold as a stand-alone component, but as part of a pre-packaged server incorporating the cooling technology.
There are no details on costs or performance so far, but some customers have already been testing prototypes, and according to Cerebras results have been promising. CEO and co-founder Andrew Feldman told Fortune that early tests show they are reducing training time from months to minutes.
We’ll have to wait until the first systems ship to customers in September to see if those claims stand up. But Feldman told ZDNet that the design of their chip should help spur greater innovation in the way engineers design neural networks. Many cornerstones of this process—for instance, tackling data in batches rather than individual data points—are guided more by the hardware limitations of GPUs than by machine learning theory, but their chip will do away with many of those obstacles.
Whether that turns out to be the case or not, the WSE might be the first indication of an innovative new era in silicon design. When Google announced it’s AI-focused Tensor Processing Unit in 2016 it was a wake-up call for chipmakers that we need some out-of-the-box thinking to square the slowing of Moore’s Law with skyrocketing demand for computing power.
It’s not just tech giants’ AI server farms driving innovation. At the other end of the spectrum, the desire to embed intelligence in everyday objects and mobile devices is pushing demand for AI chips that can run on tiny amounts of power and squeeze into the smallest form factors.
These trends have spawned renewed interest in everything from brain-inspired neuromorphic chips to optical processors, but the WSE also shows that there might be mileage in simply taking a sideways look at some of the other design decisions chipmakers have made in the past rather than just pumping ever more transistors onto a chip.
This gigantic chip might be the first exhibit in a weird and wonderful new menagerie of exotic, AI-inspired silicon.
Image Credit: Used with permission from Cerebras Systems. Continue reading
Posted in Human Robots
#435308 Brain-Machine Interfaces Are Getting ...
Elon Musk grabbed a lot of attention with his July 16 announcement that his company Neuralink plans to implant electrodes into the brains of people with paralysis by next year. Their first goal is to create assistive technology to help people who can’t move or are unable to communicate.
If you haven’t been paying attention, brain-machine interfaces (BMIs) that allow people to control robotic arms with their thoughts might sound like science fiction. But science and engineering efforts have already turned it into reality.
In a few research labs around the world, scientists and physicians have been implanting devices into the brains of people who have lost the ability to control their arms or hands for over a decade. In our own research group at the University of Pittsburgh, we’ve enabled people with paralyzed arms and hands to control robotic arms that allow them to grasp and move objects with relative ease. They can even experience touch-like sensations from their own hand when the robot grasps objects.
At its core, a BMI is pretty straightforward. In your brain, microscopic cells called neurons are sending signals back and forth to each other all the time. Everything you think, do and feel as you interact with the world around you is the result of the activity of these 80 billion or so neurons.
If you implant a tiny wire very close to one of these neurons, you can record the electrical activity it generates and send it to a computer. Record enough of these signals from the right area of the brain and it becomes possible to control computers, robots, or anything else you might want, simply by thinking about moving. But doing this comes with tremendous technical challenges, especially if you want to record from hundreds or thousands of neurons.
What Neuralink Is Bringing to the Table
Elon Musk founded Neuralink in 2017, aiming to address these challenges and raise the bar for implanted neural interfaces.
Perhaps the most impressive aspect of Neuralink’s system is the breadth and depth of their approach. Building a BMI is inherently interdisciplinary, requiring expertise in electrode design and microfabrication, implantable materials, surgical methods, electronics, packaging, neuroscience, algorithms, medicine, regulatory issues, and more. Neuralink has created a team that spans most, if not all, of these areas.
With all of this expertise, Neuralink is undoubtedly moving the field forward, and improving their technology rapidly. Individually, many of the components of their system represent significant progress along predictable paths. For example, their electrodes, that they call threads, are very small and flexible; many researchers have tried to harness those properties to minimize the chance the brain’s immune response would reject the electrodes after insertion. Neuralink has also developed high-performance miniature electronics, another focus area for labs working on BMIs.
Often overlooked in academic settings, however, is how an entire system would be efficiently implanted in a brain.
Neuralink’s BMI requires brain surgery. This is because implanted electrodes that are in intimate contact with neurons will always outperform non-invasive electrodes where neurons are far away from the electrodes sitting outside the skull. So, a critical question becomes how to minimize the surgical challenges around getting the device into a brain.
Maybe the most impressive aspect of Neuralink’s announcement was that they created a 3,000-electrode neural interface where electrodes could be implanted at a rate of between 30 and 200 per minute. Each thread of electrodes is implanted by a sophisticated surgical robot that essentially acts like a sewing machine. This all happens while specifically avoiding blood vessels that blanket the surface of the brain. The robotics and imaging that enable this feat, with tight integration to the entire device, is striking.
Neuralink has thought through the challenge of developing a clinically viable BMI from beginning to end in a way that few groups have done, though they acknowledge that many challenges remain as they work towards getting this technology into human patients in the clinic.
Figuring Out What More Electrodes Gets You
The quest for implantable devices with thousands of electrodes is not only the domain of private companies. DARPA, the NIH BRAIN Initiative, and international consortiums are working on neurotechnologies for recording and stimulating in the brain with goals of tens of thousands of electrodes. But what might scientists do with the information from 1,000, 3,000, or maybe even 100,000 neurons?
At some level, devices with more electrodes might not actually be necessary to have a meaningful impact in people’s lives. Effective control of computers for access and communication, of robotic limbs to grasp and move objects as well as of paralyzed muscles is already happening—in people. And it has been for a number of years.
Since the 1990s, the Utah Array, which has just 100 electrodes and is manufactured by Blackrock Microsystems, has been a critical device in neuroscience and clinical research. This electrode array is FDA-cleared for temporary neural recording. Several research groups, including our own, have implanted Utah Arrays in people that lasted multiple years.
Currently, the biggest constraints are related to connectors, electronics, and system-level engineering, not the implanted electrode itself—although increasing the electrodes’ lifespan to more than five years would represent a significant advance. As those technical capabilities improve, it might turn out that the ability to accurately control computers and robots is limited more by scientists’ understanding of what the neurons are saying—that is, the neural code—than by the number of electrodes on the device.
Even the most capable implanted system, and maybe the most capable devices researchers can reasonably imagine, might fall short of the goal of actually augmenting skilled human performance. Nevertheless, Neuralink’s goal of creating better BMIs has the potential to improve the lives of people who can’t move or are unable to communicate. Right now, Musk’s vision of using BMIs to meld physical brains and intelligence with artificial ones is no more than a dream.
So, what does the future look like for Neuralink and other groups creating implantable BMIs? Devices with more electrodes that last longer and are connected to smaller and more powerful wireless electronics are essential. Better devices themselves, however, are insufficient. Continued public and private investment in companies and academic research labs, as well as innovative ways for these groups to work together to share technologies and data, will be necessary to truly advance scientists’ understanding of the brain and deliver on the promise of BMIs to improve peoples’ lives.
While researchers need to keep the future societal implications of advanced neurotechnologies in mind—there’s an essential role for ethicists and regulation—BMIs could be truly transformative as they help more people overcome limitations caused by injury or disease in the brain and body.
This article is republished from The Conversation under a Creative Commons license. Read the original article.
Image Credit: UPMC/Pitt Health Sciences, / CC BY-NC-ND Continue reading
Posted in Human Robots