This is Science: What's inside a neuromorphic chip?

    After the recent announcement of a neuromorphic chip from IBM on Habré, it’s time to get acquainted with how the work of real neurons is transferred to the iron of neuromorphic chips. And the article published in ACSNano about three-dimensional electronic synapse will help us with this.

    Let's start with a little background. Once we found out where scientists get information about the device of the brain, then there was a big post with answers to readers' questions about the Human Brain project, recently an announcement of a neuromorphic chip from IBM appeared on Habré . In the series of these publications, I would also like to devote time to how neuromorphic chips are arranged at a basic level.

    One of the main components of neuromorphic calculations, in general, of neuromorphic chips, in particular, is a synapse or a system for transmitting excitation from a neuron to a neuron, because the neuron itself is often just a metal tape, a conductor. A synapse in nerve tissue is a place of contact between two neurons or between a neuron and a signal-receiving effector cell, which serves to transmit a nerve impulse.

    The main elements of the nerve cell

    By the mechanism of transmission of such an impulse, synapses can be divided into chemical - that is, using neurotransmitter molecules - and electrical - that is, due to the “breakdown” of the intercellular fluid by an electrical impulse. Electrical synapses are a pair of cell membranes located at a very close distance (only a few nm) from each other due to special proteins that perform the functions of excitation transfer.

    Usually, the data on a specific device or model of a neuromorphic chip used and the type of connections of neurons to the network in it is the know-how and subject of the NDA (not-disclosure agreement), however, in purely scientific, non-industry publications, a lot of information about their device can be found.

    So, a group of Chinese and American scientists from Peking University, Stanford and Arizona University published an article in ACSNano magazine dedicated to the development of a new three-dimensional and ultra-low-power electronic synapse, the diagram of which is shown in the figure:

    (a) A typical 2D array for an electrical neural network, where each synapse is at the intersection of the conducting lines of the pre-neuron and post-neuron. (b) The electrical circuit of the concept with synapses based on a resistive switching device. (c) The most compact layout of synapses (high density application). (d) Layout of synapses for high-accuracy computation. (e) TEM image of a cross section of the resulting electrical synapse.

    The development was based on materials with switched resistance, for example, HfO x or HfO x / AlO y , which, depending on the duration and amplitude of the applied voltage, can change their resistive properties over a wide range - potentially more than 3 orders of magnitude from 10 3 to 10 6 Ohms. This switching occurs due to the migration and rearrangement of oxygen vacancies inside the oxides.

    And in order to check the electronic synapses in the case, a two-layer neuromorphic chip was created, the first layer of which consisted of 32 x 32 neurons sensitive to pixel brightness, and the second of 16 cortex neuronsconnected by conventional or three-dimensional electrical synapses. Results on the face: a three-dimensional electronic synapse gives better recognition than usual due to a smaller deviation of resistance. In this case, the training takes place at a reduced power consumption by pulses of 50 ns, at a voltage below 2.5 V and a current below 0.3 μA

    (a) The abundance of oxygen vacancies leads to low resistance and, conversely, (b) their lack means high resistance and low current. (c) Deviation of resistance depending on energy expended. (d) The pattern used to train the system. (ef) Patterns obtained using a normal neural network and created on the basis of three-dimensional synapses, respectively. (g) Recognition accuracy.

    Compared with the primitive two-dimensional synapse neural network, the developed device looks incredibly complicated, however, the materials used are relatively cheap and ubiquitous in the electronic industry, which, according to the authors, makes it possible to produce such neuromorphic chips with extremely low cost.

    Original article in ACSNano (DOI: 10.1021 / nn501824r)

    A complete list of This is Science's published articles on GeekTimes:
    This is Science: Simple and Cheap Solar Energy
    This is Science: Graphene - Life or Death?
    This is Science: Blow and Get Electricity
    This is Science: Silicon Electronics: Bend Me All the Way!
    This is Science: An elastic display on quantum dots.
    This is Science: Putting triboelectricity at the service of humanity.
    This is Science: 3D optical printing moves to the micro level.
    This is Science: What's inside a neuromorphic chip?
    This is Science: News from graphene fields
    This is Science: 3D electronic lithography to the masses
    This is Science: Alkaline battery discharge or why the battery bounces
    This is Science: micro-guns and nanocores.
    This is Science: wearable electronics and triboelectricity. Part 1
    This is Science: Wearable Electronics and Triboelectricity. Part 2

    Sometimes it is possible to read briefly, and sometimes not so much about the news of science and technology on my Telegram channel - we are welcome;)

    Only registered users can participate in the survey. Please come in.

    What do you think, when will we see the first neuromorphic chip in desktop PCs / laptops / tablets / smartphones?

    • 3.4% Very soon, 1-2 years 23
    • 17.3% 3-5 years 117
    • 37.6% from 5-7 years 254
    • 14% We will never see, since neuromorphic chips will be used purely for supercomputer computing 95
    • 27% I - bees 182
    • 0.4% I have something to answer on the topic and write in the comments 3

    Also popular now: