Protonic Programmable Resistors for AI
Researchers at the Massachusetts Institute of Technology (MIT) have developed an analog deep learning processor based on arrayed proton programmable resistors.
In the processor, increasing and decreasing electrical conductance of proton resistors enables analog machine learning. Conductance is controlled by the movement of protons. To increase conductance, more protons are pushed into a channel of the resistor, while to decrease conductance, protons are removed. This is accomplished by using an electrolyte, similar to a battery, which conducts protons but blocks electrons.
“The device’s operating mechanism is the electrochemical insertion of the smallest ion, the proton, into an insulating oxide to modulate its electronic conductivity. Because we are working with very thin devices, we could accelerate the movement of this ion using a strong electric field and push these ion devices to the nanosecond operating regime,” said Bilge Yildiz, professor of science and engineering. nuclear and materials teacher. science and engineering at MIT.
“The action potential in biological cells rises and falls with a time scale of milliseconds, because the voltage difference of about 0.1 volts is limited by the stability of water,” said Ju Li, Professor of Nuclear Science and Engineering and Professor of Materials Science and Engineering. at MIT. “Here, we apply up to 10 volts through a special nano-thick solid glass film that conducts protons, without permanently damaging it. And the stronger the field, the faster the ion devices.
The researchers used inorganic phosphosilicate glass (PSG) for the electrolyte. PSG is similar to silicon dioxide, used in desiccant bags and in silicon processing, with a small amount of phosphorus added to give it special characteristics for proton conduction. PSG enables ultrafast proton movement because it contains a multitude of nanometer-sized pores whose surfaces provide proton scattering pathways.
Murat Onen, a postdoc at MIT, noted that it can also withstand very strong pulsating electric fields, which is critical because applying more voltage allows protons to move faster. “The speed was certainly surprising. Normally, we wouldn’t apply such extreme fields to devices, so as not to turn them to ashes. But instead, the protons ended up shuttling at immense speeds through the stack of devices, specifically a million times faster than what we had before. And this movement does not damage anything, thanks to the small size and low mass of the protons. It’s almost like teleporting.
“The nanosecond timescale means we are close to the ballistic or even quantum tunneling regime for the proton, in such an extreme field,” Li said.
Because protons do not damage the material, the resistor can operate for millions of cycles without failing. Onen added that the insulating properties of PSG mean that almost no electric current passes through the material as the protons travel, making the device extremely energy efficient. It is also compatible with silicon manufacturing techniques.
The researchers plan to redesign the programmable resistors for large-scale manufacturing, then study the properties of resistor networks and scale them up so they can be integrated into systems. They also plan to study materials to eliminate bottlenecks that limit the voltage needed to efficiently transfer protons to, through, and from the electrolyte.
“Another exciting direction that these ion devices can enable is energy-efficient hardware to emulate neural circuits and synaptic plasticity rules that are inferred in neuroscience beyond analog deep neural networks,” Yildiz added.
Optical associative learning
Researchers from the University of Oxford, University of Exeter and University of Munster have developed an on-chip optical processor capable of detecting similarities in data sets up to 1,000 times faster than conventional machine learning algorithms running on electronic processors.
The work is inspired by Pavlov’s famous classical conditioning experiments.
“Pavlovian associative learning is considered a basic form of learning that shapes the behavior of humans and animals – but adoption in AI systems is largely unknown. Our research on Pavlovian learning in tandem with optical parallel processing demonstrate the exciting potential of a variety of AI tasks,” said James Tan You Sian, from the Department of Materials, University of Oxford.
Instead of relying on backpropagation, which neural networks use to refine results, the team’s Monadic Associative Learning Element (AMLE) uses memory hardware that learns patterns to associate similar features. in datasets, mimicking the conditional reflex observed by Pavlov in the case of a match.
AMLE inputs are paired with the correct outputs to supervise the learning process, and the memory hardware can be reset using light signals. In testing, AMLE was able to correctly identify cat/non-cat images after being trained with only five image pairs.
The chip uses wavelength division multiplexing to send multiple optical signals at different wavelengths over a single channel to increase computing speed.
An associative learning approach could complement neural networks rather than replace them, noted Zengguang Cheng, a professor currently at Fudan University. “It’s most effective for problems that don’t require substantial analysis of very complex features of datasets. Many learning tasks are volume-based and do not have this level of complexity – in these cases, associative learning can accomplish the tasks faster and at lower computational cost.
Quantum comparative analysis of the anti-butterfly effect
Researchers at Los Alamos National Laboratory propose a new method to compare the performance of quantum computers.
“Using the simple and robust protocol we have developed, we can determine how efficiently quantum computers can process information, and this also applies to information loss in other complex quantum systems,” said said Bin Yan, a quantum theorist at Los Alamos National. Laboratory. “Our protocol quantifies information scrambling in a quantum system and unambiguously distinguishes it from false positive signals in the background noise caused by quantum decoherence.”
Decoherence erases all quantum information from a quantum computer, while scrambling information through quantum chaos diffuses information throughout the system, protecting it and allowing it to be recovered.
“Our method, which builds on the quantum butterfly anti-butterfly effect we discovered two years ago, evolves a system forward and backward in time in a single loop, so that we can apply it to any system that has time-reversing dynamics, including quantum computers and quantum simulators using cold atoms,” Yan said.
The team demonstrated the protocol with simulations on cloud-based IBM quantum computers. The researchers explain that the method prepares a quantum system and subsystem, evolves the complete system forward in time, causes a change in a different subsystem, and then evolves the system backwards during the same duration. The measurement of information overlap between the two subsystems shows how much information has been preserved by scrambling and how much information has been lost by decoherence.
#Research #Pieces #December