The artificial intelligence (AI) program AlphaGo uses 1200 CPUs and 56,000 watts per hour. This is very different from humans who only have one CPU and use 20 watts when being asked to perform a similar task. Professor YU Woo Jong notes that AI will use a lot less power if hardware can be developed that is based on the human brain’s structure.
Inspired by the neuron connections of the human brain, a joint team from the Center for Integrated Nanostructure Physics within the Institute for Basic Science (IBS), and Sungkyunkwan University, have developed a new memory device. In addition to having endurance, highly reliable performance and a long retention time, the device also boasts flexibility and stretch-ability. The research was published in Nature Communications and notes that these characteristics makes the device promising tool for the next-generation soft electronics that could be attached to the body or clothes.
Information memorized in the brain is transmitted as an electro-chemical signal through synapses from one neuron to the next. Due to a huge number of connections between neurons, the human brain can memorize and learn. IBS scientists, using these connections as an example, created a memory called two-terminal tunneling random access memory (TRAM). In TRAM, the two communicating neurons of the synapse are mimicked by two electrodes, known as the drain and source. Although most mobile electronics like mobile phones and digital cameras use three-terminal flash memory, two-terminal memories like TRAM have an advantage in that they do not need a thick and rigid oxide layer. Professor Yu notes that TRAM is more flexible and can be scalable, but admits that flash memory is still more reliable and has better performance.
A few atom-thick 2D crystal or a stack of one-atom-thick layers is used to create TRAM. The first layer with the drain and source electrodes is made from the semiconductor molybdenum disulfide (MoS2). Hexagonal boron nitride (h-BN) is used as an insulating layer and the final layer is made of graphene. Memory is created, read or erased by the flow of current through these layers. Data storage is achieved by electrons being kept on the graphene layer. Different voltages are used between the electrodes. This causes electrons to flow from the drain to the graphene layer by tunneling through the insulating h-BN layer. Memory is written and stored when the graphene layer is negatively charged and erased when positive charges are introduced.
Experimentation showed that a thickness of 7.5 nanometers of the insulating h-BN layer is the most effective for allowing electrons to tunnel to the graphene layer without losing flexibility, but also without leakages.
TRAM’s two key features are stretch-ability and flexibility. TRAM was manufactured on stretchable silicone materials (PDMS) and flexible plastic (PET). It was found that it could be strained up to 20% and 0.3% respectively.
Future applications for TRAM include saving data from eye cameras, flexible or wearable smartphones, body-attachable biomedical devices and smart surgical gloves.
TRAM’s performance is better than other types of two-terminal memories known as resistive random-access memory (RRAM) and phase-change random-access memory (PRAM).