Power/Performance Bits: Oct. 15

Probabilistic computing; memory compaction for C; ear biometrics.


Probabilistic computing
Researchers at Purdue University and Tohoku University built a hardware demonstration of a probabilistic computer utilizing p-bits to perform quantum computer-like calculations. The team says probabilistic computing could bridge the gap between classical and quantum computing and more efficiently solve problems in areas such as drug research, encryption and cybersecurity, financial services, data analysis and supply chain logistics.

P-bits were proposed in 2017 and can be either zero or one at any given time and fluctuate rapidly between the two.

“There is a useful subset of problems solvable with qubits that can also be solved with p-bits. You might say that a p-bit is a ‘poor man’s qubit,’” said Supriyo Datta, a Professor of Electrical and Computer Engineering at Purdue who led the group that proposed them.

Unlike qubits, which require extremely cold temperatures, p-bits work at room temperature.

The device uses magnetoresistive RAM (MRAM) that was modified to make it intentionally unstable, allowing p-bits to fluctuate. This was combined with a transistor to build a three-terminal unit whose fluctuations could be controlled. Eight such p-bit units were interconnected to build a probabilistic computer.

For the first time, researchers have demonstrated a way to build a probabilistic computer. This circuit includes a modified version of a magnetoresistive random-access memory device (red) to interconnect eight p-bits. (Credit: Purdue University image/Ahmed Zeeshan Pervaiz)

The circuit was able to solve integer factorization calculations, often considered a ‘quantum’ problem. While classical computers are capable of integer factorization, the researchers said the probabilistic approach can solve them with much less space and energy.

“On a chip, this circuit would take up the same area as a transistor, but perform a function that would have taken thousands of transistors to perform. It also operates in a manner that could speed up calculation through the parallel operation of a large number of p-bits,” said Ahmed Zeeshan Pervaiz, a Ph.D. student in electrical and computer engineering at Purdue.

While hundreds of p-bits would be needed to solve more complex problems, the team says creating such a device is not far off.

Memory compaction for C
Researchers at the University of Massachusetts Amherst propose a way to reduce the memory demands of C, C++, and Objective C applications. The system, called Mesh, is a plug-in replacement for malloc that uses virtual memory operations to reduce fragmentation.

Memory fragmentation is a big problem for programs written in C-like languages, said Emery Berger, a professor of computer science at the University of Massachusetts Amherst. “This is how memory gets wasted. Imagine a Tetris board where you could stop and reorganize it at any time – this would make the game a lot easier, because you could always squeeze out the empty space. But you can’t do this in C, just as you can’t do it in Tetris.”

Mesh uses virtual memory to perform memory compaction without changing the addresses of objects.

“The trick is to find chunks of memory that can be interleaved, sort of like when interlocking gears mesh,” Berger explained.

Mesh manages memory in spans of 4K contiguous pages. Each span only contains same-sized objects and randomized allocation distributes objects uniformly at random across all available offsets in a span.

Mesh then searches for spans that are candidates for meshing, where utilization is low and allocations are at non-overlapping offsets. Virtual memory is then updated so both virtual pages point to the same physical page.

“This meshing process works because we only change things in ‘physical’ memory. From the perspective of the program, which can only see ‘virtual’ memory, nothing has changed. This is powerful because we can do this for any application automatically.”

In tests, Mesh was able to automatically reduce the memory demands of the Firefox web browser by 16% while only impacting runtime performance by 1%. The team also applied Mesh to Redis, a popular open source data structure server, which reduced memory demands by 39%.

Ear biometrics
Researchers at the University at Buffalo and Syracuse University propose a new way to unlock your smartphone: ear biometrics. Inspired by the ubiquity of earbuds on streets and college campuses, the researchers started on a project that uses modified earbuds to authenticate smartphone users via the unique geometry of their ear canal.

“We have so many students walking around with speakers in their ears. It led me to wonder what else we could do with them,” said Zhanpeng Jin, associate professor in the Department of Computer Science and Engineering at UB.

A prototype of the system, called EarEcho, was created using off-the-shelf in-ear headphones and a tiny microphone. The team developed acoustic signal processing techniques to limit noise interference, and models to share information between EarEcho’s components.

When a sound is played into someone’s ear, the sound propagates through and is reflected and absorbed by the ear canal, producing a unique signature that can be recorded by the microphone. The signature is sent by the earbuds’ Bluetooth to the connected smartphone for analysis.

“It doesn’t matter what the sound is, everyone’s ears are different and we can show that in the audio recording,” says Jin. “This uniqueness can lead to a new way of confirming the identity of the user, equivalent to fingerprinting.”

To test the device, 20 subjects listened to audio samples that included a variety of speech, music and other content. The team conducted tests in different environmental settings (on the street, in a shopping mall, etc.) and with the subjects in different positions (sitting, standing, head tilted, etc.).

EarEcho was roughly 95% effective when given 1 second to authenticate the subjects. The score improved to 97.5% when it continued to monitor the subject in 3 second windows.

While EarEcho could be used to unlock smartphones, the team sees the most value in continual monitoring, where a user wouldn’t need to verify their identity for making mobile payments, for example, or other tasks that need fingerprint or password authentication. “Just by wearing the earphones, which many people already do, you wouldn’t have to do anything to unlock your phone,” Jin said.

Leave a Reply

(Note: This name will be displayed publicly)