HP Labs

HP Labs successfully demonstrates a path towards “Physical Computing”

By Simon Firth, HP Labs Correspondent — January 5, 2017

 

Master Technologist Ning Ge and Distinguished Technologist Helen Holder

Master Technologist Ning Ge and Distinguished Technologist Helen Holder

A paper published today in Nature: Scientific Reports details the demonstration at HP Labs of a new kind of computing system: a type of “physical computer” that inputs, manipulates, and stores information without the original analog data being converted into the digital ones and zeros that are at the heart of conventional computing.

The research is the result of a collaboration between HP Labs, Hewlett Packard Labs, the University of Massachusetts at Amherst (UMASS) and Nanyang Technological University in Singapore and holds out the promise of a new generation of highly energy efficient, low cost electronic devices that could enable the vast sensor networks imagined as the basis for “ambient” computing – a future ‘megatrend’ being investigated in HP’s Emerging Compute Lab.

“While the idea of a physical computer is not new, we’ve been able show that memristor technology developed at HP Labs can be used to move physical computing in a new direction,” says Ning Ge, an HP Labs Master Technologist and co-lead author of the paper. “It’s a pretty encouraging result.”

“Ambient computing demands computational capability anywhere, anytime,” adds Helen Holder, who leads Emerging Compute Lab nano-scale research. “This paper suggests a route that might take us where we need to go.”

The task of converting an analog input (be it text, audio, video, or any kind of scientific measurement) into digital information before it can be manipulated is becoming a major issue for traditional computers. Specialized chips called accelerators typically do this work, but they are governed by the same physical laws that suggest all transistors are reaching the point where they cannot be made physically smaller without inducing unpredictable quantum effects.  

As a result, researchers have looked to develop other computing paradigms that could absorb ever more data ever faster, including quantum, neuromorphic, and biological computing. In the last decade, HP Labs has pioneered a fourth paradigm built around memristors – a non-volatile electrical component that ‘remembers’ the electrical resistance of the current that most recently flowed through it – often described as memory-driven computing.

A former member of HP Lab’s memristor team, Ge realized that memristors could also be used to tackle one of the main jobs in computing: noting the differences between any two sets of data.

“That’s essential for all kinds of tasks in modern computer and communication systems,” notes Ning. “We use it for something as simple as comparing temperatures from one minute to the next in small sensing devices , for password checking on your phone, or for establishing whether two sets of millions of lines of code are perfect copies of each other or not.”

Crucially, memristors can make these comparisons without needing to first convert an analog input into digital information. In their experiment, Ge and his colleagues showed that a memristor array designed at HP and built by Professor Joshua Yang’s research group at UMASS could receive two voltage readings from the physical world and record the difference between them in one step without their ever being converted into binary code. Moreover, the results of that comparison were stored in the non-volatile system without requiring energy to maintain them in memory.

“Our paper describes an architecture for doing these comparisons in a way that is both simple and elegant and is much more efficient than conventional comparators,” says Ge.

The work holds out hope of fast and power-thrifty memristor-based accelerators that could take over the comparative work we currently ask of conventional digital accelerators. That would help conventional digital CPUs keep up with higher demand even as they brush against their physical limits.

In comparing two inputs and establishing which most closely matches a pre-established reference point, these novel comparators are also performing one of the basic functions of computing, suggesting they could act as building blocks for more complex kinds of physical computers.

More immediately, this research offers a path toward the development of electronic devices that enable ambient computing, a successor to the concept of an Internet of Things and an area of research being pioneered at HP Labs.

“The future office, for example, will have many thousands of small, networked sensors that monitor temperature, light, and human presence to optimize the building’s energy use,” Ge explains. “Physical computing devices can make that vision much easier and cheaper to realize.”