SoylentNews

arulatas writes:

A new way of creating a neural network using specially formulated memristors has been described by a team of researchers from Stony Brook University and the University of California Santa Barbara. The process has the potential to place an entire neural network on a single chip:

The system produced by the authors here involved only a 12-by-12 grid of memristors, so it's pretty limited in capacity. But Robert Legenstein, from Austria's Graz University of Technology, writes in an accompanying perspective that "If this design can be scaled up to large network sizes, it will affect the future of computing."

That's because there are still many challenges where a neural network can easily outperform traditional computing hardware—and do so at a fraction of the energy cost. Even on a 30 nm process, it would be possible to place 25 million cells in a square centimeter, with 10,000 synapses on each cell. And all that would dissipate about a Watt.

Training and operation of an integrated neuromorphic network based on metal-oxide memristors [abstract]

MIT Technology Review

takyon writes:

The Next Platform has an article about waning interest in brain-inspired neuromorphic computing post-2013 (which has not yet delivered a "revolution in computing") and some of the developments in the field since then:

There have been a couple of noteworthy investments that have fed existing research for neuromorphic architectures. The DARPA Synapse program was one such effort, which beginning in 2008, eventually yielded IBM's "True North" chip—a 4096-core device comprised of 256 programmable "neurons" that act much like synapses in the brain, resulting in a highly energy efficient architecture that while fascinating—means an entire rethink of programming approaches. Since that time, other funding from scientific sources, including the Human Brain Project, have pushed the area further, leading to the creation of the SpiNNaker neuromorphic device, although there is still a lack of a single architecture that appears best for neuromorphic computing in general.

The problem is really that there is no "general" purpose for such devices as of yet and no widely accepted device or programmatic approach. Much of this stems from the fact that many of the existing projects are built around specific goals that vary widely. For starters, there are projects around broader neuromorphic engineering that are more centered on robotics versus large-scale computing applications (and vice versa). One of several computing-oriented approaches taken by Stanford University's Neurogrid project, which was presented in hardware in 2009 and remains an ongoing research endeavor, was to simulate the human brain, thus the programming approach and hardware design are both thus modeled as closely to the brain as possible while others are more oriented toward solving computer science related challenges related to power consumption and computational capability using the same concepts, including a 2011 effort at MIT, work at HP with memristors as a key to neuromorphic device creation, and various other smaller projects, including one spin-off of the True North architecture we described here.

[more]

takyon writes:

New quick-learning neural network powered by memristors

A new type of neural network made with memristors can dramatically improve the efficiency of teaching machines to think like humans. The network, called a reservoir computing system, could predict words before they are said during conversation, and help predict future outcomes based on the present.

The research team that created the reservoir computing system, led by Wei Lu, U-M professor of electrical engineering and computer science, recently published their work [open, DOI: 10.1038/s41467-017-02337-y] [DX] in Nature Communications.

Reservoir computing systems, which improve on a typical neural network's capacity and reduce the required training time, have been created in the past with larger optical components. However, the U-M group created their system using memristors, which require less space and can be integrated more easily into existing silicon-based electronics. [...] When a set of data is inputted into the reservoir, the reservoir identifies important time-related features of the data, and hands it off in a simpler format to a second network. This second network then only needs training like simpler neural networks, changing weights of the features and outputs that the first network passed on until it achieves an acceptable level of error. "The beauty of reservoir computing is that while we design it, we don't have to train it," says Lu.

[...] Using only 88 memristors as nodes to identify handwritten versions of numerals, compared to a conventional network that would require thousands of nodes for the task, the reservoir achieved 91% accuracy.

Original Submission

NPC-131072 writes:

From IEEE Spectrum:

Hoping to speed AI and neuromorphic computing and cut down on power consumption, startups, scientists, and established chip companies have all been looking to do more computing in memory rather than in a processor's computing core. Memristors and other nonvolatile memory seem to lend themselves to the task particularly well. However, most demonstrations of in-memory computing have been in standalone accelerator chips that either are built for a particular type of AI problem or that need the off-chip resources of a separate processor in order to operate. University of Michigan engineers are claiming the first memristor-based programmable computer for AI that can work on all its own.

Original Submission