Microelectronics technology has changed our way of life.

Our way of life, in turn, is now triggering fundamental changes in microelectronics.

Microelectronics technology provides us with the capability to generate, process and store huge amounts of data. A large proportion of this data is no longer “structured” in databases or spreadsheets. It is “unstructured” and cannot be easily grasped by today’s computing systems. Consider, for example, the need to interpret text, images or sounds. Consequently, we need new tools to cope with the large amounts of data created by microelectronics technology and to extract useful information from that data. Our society needs machines that can recognize complex patterns, deliver cross-correlated information and, eventually, learn on their own. They will proceed from large datasets or from integrated sensors that can adapt to their environment and extract the most valuable information. Neural networks, inspired by the mechanisms in our brain, are well suited to interpret unstructured data. As a result, there is a need for “neuromorphic processing units” to become the new heart that powers such machines.

We need new tools to cope with the large amounts of data cre­ated by micro­elec­tron­ics tech­nol­ogy.

—IBM scientist Bert Jan Offrein

Our mission is to

  • innovate with materials that enable the processing of electronic and photonic signals in brain-inspired circuits,
  • invent the technology to fabricate such circuits and
  • demonstrate power-efficient hardware solutions.

Focus areas

Neuromorphic processing units require devices with plasticity, as in biology where the strength of connections between neurons in the brain is altered based on the timing of a neuron’s input and output action potentials. In order to achieve artificial electrical or optical neural networks, similar functionality is required. Connection strengths must be altered in a controlled way by applying an input signal, and the state of the device should depend on its history. We are focusing on the follow two aspects to achieve synaptic function.

First, we are developing new materials — in particular transition metal oxides — whose intrinsic properties provide plasticity at the atomic scale. For this we are concentrating on three different physical phenomena to fabricate electronic synapses:

  • Oxygen ions, which can be incorporated into an oxide matrix to continuously change its electrical resistance;
  • Ferroelectric domains, which can be reconfigured in a tunnel barrier to alter its electrical conductance; and
  • Charge carriers, which can be displaced by a light pulse to modify the refractive index locally.

Exploiting these physical effects into a neuromorphic processing unit requires a way to integrate many devices in complex circuits. It also requires a vision to co-integrate novel devices with CMOS or integrated photonic technology. Our second focus is to develop methods that enable this vision, from materials deposition and processing to direct wafer bonding, and from single device processing to post-processing arrays over CMOS circuits. Our goal is to build small demonstrators as proof of concept, for example by showing circuits emulating neurons connected by an array of electronic synapses.

Ask the experts

Bert Jan Offrein

Bert Jan Offrein

IBM Research scientist

Jean Fompeyrine

Jean Fompeyrine

IBM Research scientist

Stefan Abel

Stefan Abel

IBM Research scientist


EU projects

ULPEC logo

ULPEC

For an Ultra-Low Power Event-Based Camera


NeuRAM3 logo

NeuRAM3

NEUral computing aRchitectures in Advanced Monolithic 3D-VLSI nano-technologies


NeuRAM3 logo

PHRESCO

PHotonic REServoir COmputing