Cool. Lithium-ion batteries can be used as hard drives.

If lithium-ion batteries had anything to do with computers, would it surprise you if I said that lithium-ion batteries could now be used as an important electronic component in computer high-precision circuits? Yes, when the editor first got this article, he couldn't believe his own eyes. Lithium-ion batteries, which seem to lack the technology to store energy, can actually be used in the high-precision circuits of electronic computers. It is really confusing. There is only one sentence in the heart of the editor relative to the lithium ion battery: Why don't you go to heaven?

In recent years, with the rapid development of computer technology, especially the development of artificial neural network technology, the demand for large-scale information storage has become more and more intense, but traditional storage components, For example, SRAM and DRAM use a logical architecture for serial operations in the storage process, so it not only leads to low storage efficiency, but also greatly increases the energy consumption in the information storage process. In order to overcome the shortcomings of traditional memory, our respectable engineers developed cross-memory. The remarkable feature of this memory is the use of parallel operation logic architecture, which improves the efficiency of memory and reduces the energy consumption of information storage.

Recent reports have shown that neural network prototypes based on cross-array structures of filamentous metal oxide memristors have been trained to perform simple graphic identification and file classification. However, there are still many limitations in these memories, which affect the correctness and energy consumption efficiency of these storage components. The first one is serious writing noise, writing nonlinear and excessively high switching voltage and current.

In order to overcome the current limitations of cross-storage, EliotJ of the Sandia National Laboratory in the United States. Fuller developed a transistor based on a lithium-ion battery structure(LISTA), which is a all-solid non-volatile Redox resistance switch transistor. The mechanism of action is to use the voltage applied to both ends of LISTA to drive lithium ions to embed and exit between positive and negative poles to achieve the purpose of controlling transistor switches. Compared to other types of memory, the energy barrier between LISTA's solid electrolytes and conductive channels is very low. The activation energy of Li + diffusion in LiCoO2 is only about 0.25 eV, Li + diffuses in LiCoO2, and under low current conditions, The minimum overpotential is only 5 mV, and it is only 100 mV under the large current, so only a very small voltage is required to control its switching state. More importantly, in the course of LISTA's operation, there will be no large phase transitions like other memories, and the diffusion of lithium ions in LiCoO2 will only slightly change its crystal structure. Therefore, compared to other types of memory, LISTA has the advantages of low write noise, linear operation and low operating voltage and current, which can greatly reduce memory power consumption and is very suitable for application in artificial neural networks.

The structure of LISTA is shown above. It is covered with a layer of Pt and leakage electrodes on the substrate of SiO2 and is connected to a layer of 120nm-thick LiCoO2 conductive pathway with a 400nm-thick LPON solid electrolyte in the middle and a layer on the other side. A 20nm-thick Sigate, The working principle is to drive Li + to transfer from the conductive pathway to the gate electrode by applying a voltage on the gate electrode, thereby controlling the concentration of Li + in the conductive pathway, and during the process of Li + migrating out of LiCoO2, LiCoO2 completed the transition from an insulator to a metal conductor to achieve the purpose of controlling the conductivity of the conductive path. The main reason for choosing LiCoO2 is its good stability in solid electrolytes and its good recycling performance. LISTA's switching state can be maintained for weeks or even months. Although the switch will eventually be turned off due to self-discharge, it can be well controlled by increasing the diffusion barrier. In fact, the current technology can already control LISTA's self-discharge rate at a level of <UNK> 3 <UNK> / year, and neural networks generally do not require storage time of up to 10 years. Generally, information in the storage network is rewritten every week. Therefore, the current performance of LISTA can already meet existing requirements. The reading and writing test showed that at a capacity of 10 <UNK>, LISTA has a life span of 100,000 times, which basically meets the needs of artificial neural networks, and can further improve the recycling performance of LISTA by increasing the crystal degree of the positive pole and reducing the thickness. Moreover, LISTA has the highest signal-to-noise ratio in current storage components and is very suitable for application in artificial neural networks.

EliotJ. Fuller's work has created a new idea for the development of low-power, low-noise, and high-linear memory required by artificial neural networks, using Li +'s advantages of low spread barriers and low activation energy in LiCoO2. A high-performance memory LISTA with extremely low power consumption was developed, which greatly improved the energy consumption efficiency of large-scale information storage, contributed to the development of artificial neural networks, and was also another major contribution of lithium-ion batteries to the development of human science and technology.

References:

Li-Ion Synthetic Transform Lowe Power Analyzing, Adv. Mater. 2016, ElliotJ. Fuller, et. al

The page contains the contents of the machine translation.