Breaking through a technological roadblock that has long limited efficient edge-AI learning, a team of French scientists developed the first hybrid memory technology to support adaptive local training and inference of artificial neural networks.

In a paper titled "A Ferroelectric-Memristor Memory for Both Training and Inference" published in Nature Electronics, the team presents a new hybrid memory system that combines the best traits of two previously incompatible technologies—ferroelectric capacitors and memristors—into a single, CMOS-compatible memory stack. This novel architecture delivers a long-sought solution to one of edge AI's most vexing challenges: how to perform both learning and inference on a chip without burning through energy budgets or challenging hardware constraints.

Memristors are a fourth class of electric circuit, first hypothesized way back in the 1970's, which are a new addition to the transistors, capacitors, resistors etc that go into making a silicon chip. HP has put a great deal of resources into developing this technology and is expected to release memory-like memristor products in 2013, so it's not far off from commercial deployment. Now however, the Korean Advanced Institute of Science and Technology (KAIST) have found a way to use memristors to make what they describe as a "fully functional" flexible resistive random access memory (RRAM). This memory is built on a plastic substrate and can be randomly accessed, written to and erased. However, as this substrate is flexible, it can be bent and rolled up easily, opening up possibilities for use that haven't even been thought of yet.

HP today announced that it has entered into a joint development agreement with Hynix Semiconductor Inc., a world-leading memory supplier, to bring memristor, a new circuit element first intentionally demonstrated in HP Labs, to market in future memory products. Memristors require less energy to operate, are faster than present solid-state storage technologies and can retain information even when power is off. The memristor, short for "memory resistor," was postulated to be the fourth basic circuit element by Prof. Leon Chua of the University of California at Berkeley in 1971 and first intentionally reduced to practice by researchers in HP Labs, the company's central research arm, in 2006.

Earlier this year, HP announced the discovery that the memristor also can perform logic, showing that memristor-based devices could change the standard paradigm of computing by enabling computation to one day be performed in chips where data is stored, rather than on a specialized central processing unit.

HP today announced that researchers from HP Labs, the company's central research facility, have proven the existence of what had previously been only theorized as the fourth fundamental circuit element in electrical engineering. This scientific advancement could make it possible to develop computer systems that have memories that do not forget, do not need to be booted up, consume far less power and associate information in a manner similar to that of the human brain.