On August 3, 2016 (July 1, 2016 lunar calendar), IBM invented the world's first artificial nerve cell artificial intelligence cornerstone has been completed. Image source: IBM Editor's note: After AlphaGo defeated Lee Sedol and announced that the supercomputer had conquered the impossible field of Go, artificial intelligence (AI) has become one of the hottest topics for everyone. For many scientists in the field, one of the ultimate goals of artificial intelligence is to use machines to achieve all the functions of the human brain, and as the smallest cell unit of the human brain, nerve cells may be the best place to start. On August 3, 2016, local time in the United States, IBM officially announced their latest achievement - the first artificial nerve cell, which can be used to create high-density, low-power cognitive learning chips. IBM Zurich Research Center has made the world's first artificial nanoscale random phase-change nerve cell. IBM has built an array of 500 of these nerve cells and made the array process signals in a way that mimics the working of the human brain. This technological breakthrough is significant because the phase-change nerve cell has properties that traditional materials cannot match - its size can be as small as nanometers. In addition, its signal transmission speed is fast and power consumption is very low. What's more, phase-change nerve cells are random, which means that under the same input signal, the output of multiple phase-change nerve cells will be slightly different, which is the property of biological nerve cells. First author of the artificial nerve cell paper: Thomas Tuma IBM Phase-change nerve cells consist of an input (dendrites similar to biological nerve cells), a neural membrane (bilayer similar to biological nerve cells), a signal generator (body of nerve cells similar to biological nerve cells), and an output (axons similar to biological nerve cells). There are also feedback loops between the signal generator and the input to enhance certain types of input signals. Artificial nerve cell R & D team, Image credit: IBM Neural membrane is the key to the entire nerve cell. In biological nerve cells, the neural membrane functions as a liquid membrane, and its TCE-metal mechanism is similar to resistance and capacitance: it prevents current from passing directly through, but at the same time absorbs energy. When the energy is absorbed to a certain extent, it emits its own generated signal. This signal is conducted along the axon and received by other nerve cells. The process is then repeated. In the nerve cell manufactured by IBM, the liquid membrane is replaced by a small piece of nerve membrane. Neural thin films are made of germanium-antimony-tellurium composites (also known as GST materials), which are also the main functional materials for rewritable Blu-ray discs. The germanium-antimony-tellurium composite is a phase change material, that is, it can exist in two states: a crystalline state and an amorphous state. The two states can be converted to each other by providing energy through a laser or an electric current. In different states, the physical properties of the phase change material are very different: the germanium-antimony-tellurium composite does not conduct electricity in the amorphous state, but conducts electricity in the crystalline state. In the artificial nerve cell, the germanium-antimony-tellurium film is initially amorphous. As the signal arrives, the Eventually, an electric current passes through the membrane, creating a signal that is emitted through the output of the nerve cell. After a certain period of time, the germanium, antimony, and tellurium film reverts to an amorphous form. The process repeats itself. Comparison of biological and artificial nerve cells, image source: IBM Biological nerve cells are stochastic due to the presence of various noises in living organisms. The IBM researchers said that artificial nerve cells also exhibited random characteristics, because the state of the nerve cell film was slightly different after each reset, so the subsequent crystallization process was slightly different. Therefore, scientists cannot know exactly what signal the artificial nerve cell will emit each time. So what is the significance of artificial nerve cells? First, artificial nerve cells use mature materials, go through billions of jobs without damage (long life), and are extremely small (some reports say 90 nanometers, but it should be around 300 nanometers from the figure below, and the paper says it is expected to reach 14 nanometers in the future). Therefore, this is a very good device. Artificial nerve cell network. The silver square in the picture is the enlarged phase-change nerve cell, which is not yet equipped with an industry-standard input-output interface. Image source: IBM Second, artificial nerve cells work very similar to biological nerve cells. When large numbers of artificial nerve cells form a parallel computing machine, it may be able to make decisions and process sensory information like humans. IBM says its artificial nerve cell technology complements another type of artificial nerve cell device currently in development, the memristor. Currently, IBM makes 10-by-10 nerve cell arrays, combining five small arrays into a large array of 500 nerve cells that can process signals in a way similar to the human brain. In fact, artificial nerve cells already exhibit the same "collective coding" properties as human nerve cells. In addition, its signal processing power has exceeded the limits stipulated by Nyquist-Shannon's sampling theorem. Editor's Note: Collective coding: Each nerve cell has 2 states, which can represent 1 bit of information, so N neurons can represent 2N bits of information. When there are enough nerve cells, the amount of information that can be represented can be extremely staggering. IBM researchers plan to build a single chip containing several thousand phase-change nerve cells and write software that can take full advantage of the random properties of phase-change nerve cell chips.