November 05, 2020
RRAM resistive memory (RRAM) based on bistable resistance conversion, as a general-purpose memory integrating dynamic/static random access memory and floating gate memory functions, is the next generation of nonvolatile memory (NVM) technology after NAND technology.
At present, the industry generally believes that RRAM is the leader in the next round of non-volatile memory competition, which is derived from the rich physical properties of resistive memory materials and their unique storage principles. Resistive memory materials include oxide materials, chalcogenide semiconductor materials, 2D materials, and organic polymer materials. According to the storage principle, these materials can be used to construct redox memories, ferroelectric tunnel junction memories, Mott memories, molecular memories and flexible memories. Among them, some unique storage principles and functions enable these memories to be extended to multi-link applications, showing their talents in multi-field intersection.
(1) Redox resistive memory and its materials. This type of memory includes two types, namely: EMB or CBRAM (electrochemical metallization bridge memory, some also called programmable metallization unit PMC) based on the behavior of metal cations and VCM (Valence Change Memory) based on oxygen anions. Type memory).
EMB resistive memory usually adopts an active electrode-ion conductor-inert electrode system. Electrochemically active materials such as Ag or Cu can be made into active electrodes. Inert electrodes are made of W or Pt. The ion conductor can be a solid electrolyte film (such as silver sulfur Series compounds Ag2S and Cu2S, etc.), can also be oxides (such as ZrO2). The formation of conductive filaments or metalized bridges depends on the electrochemical reaction of active metals. This kind of RAM technology can meet various market demands such as solid-state hard drives and embedded non-volatile memories. In 2012, commercial products of serial non-volatile memory based on EMB have come out.
In February 2015, Micron released a 16Gb CuCBRAM chip manufactured with 27nm technology (see Figure 4). In addition, EMBRRAM has attracted much attention for its reconfigurable switches for electrochemical reaction regulation. Not only can it be used as a synaptic element in artificial neural network technology (see Figure 5), but also the integration with other systems can provide more abundant Features. For example, in 2014, a supercapacitor system integrated with a resistive memory was reported, and the stability of the discharge process was greatly improved.
VCM resistive memories that rely on oxygen anions (or oxygen vacancies) are further divided into metal oxide bipolar (MOBF) RRAM and metal oxide unipolar (MO-UF) RRAM according to the resistance state transition behavior. The former generally adopts a double-layer oxide structure, in which at least one layer is a non-stoichiometric oxide film, such as the commercially available Ta2O5-x/TaO2-x structure. In 2012, Panasonic demonstrated an 8Mb TaOxRRAM with a write pulse speed of 8.2ns and a rate of 443Mb/s. In early 2013, Toshiba released a 32Gb RRAM memory integrated with 24nm CMOS.
In July of the same year, Panasonic introduced the first commercial 8-bit microcontroller to replace EEPROM, which uses a 0.18μm CMOS processor with integrated TaOxRRAM. Another type of MO-UFRRAM is also called thermochemical memory (TCM). The dielectric materials used are NiOx, HfOx, etc., and the electrodes are generally TIN, Pt, and W. Different from the foregoing storage structure, MO-UF devices do not rely on structural asymmetry to achieve resistive storage, and their write/erase voltage polarities are the same, so they have advantages in preparing small-sized memory arrays. In addition, its single voltage polarity programming method also greatly simplifies the circuit.
(2) Ferroelectric tunnel junction (FTJ) memory and its materials. A common ferroelectric tunnel junction uses a sandwich structure of metal/ultra-thin ferroelectric film/metal, and a ferroelectric layer of a few nanometers thick acts as an electron tunneling barrier. Its spontaneous polarization inversion makes the barrier height change significantly. Two resistance states, high and low, are obtained in the tunnel junction.
The fast flipping ability of ferroelectric polarization makes the ferroelectric tunnel junction resistive memory unique in terms of speed and low power consumption. For example, its read voltage is usually only 100mV, and the energy of write operations can be reduced to 10fJ/bit. FTJ encountered a technical bottleneck due to its low tunneling resistance switch ratio. Later, it surpassed this obstacle by cleverly designing a ferroelectric heterostructure. In 2013, a new structure of BaTIO3FTJ memory was reported, which included a metal electrode at one end. Substituting ferroelectric semiconductors, so as to realize the simultaneous regulation of the height and width of the barrier by the spontaneous polarization inversion of the ferroelectric, and obtain an abnormally enhanced tunnel junction resistance switch ratio. Moreover, the resistance state of such ferroelectric heterostructures depends on the ferroelectric domain structure, and a controllable quasi-continuous resistance state change can be achieved through microstructure regulation, which is a key link to realize artificial cognition. FTJ can perform functions such as "brain-like" coding, training, and recognition.
(3) Mott memory and its materials. Using Mott insulator material to produce a metal state-insulation state transition under external stimuli (light, heat, electricity) to achieve bistable storage. VO2 and NbO2 are representative materials among them. These materials not only have the ability of stimulus-dependent resistance regulation, but also can obtain stimulus-dependent capacitance regulation, so they can perform multi-link, cross-domain complex tasks. The resonator with instant adjustable resonant frequency reported by "Science" in 2009 just made use of the VO2Mott memory container's ability to regulate capacitance.
In 2014, UC Berkley of the United States reported a twistable artificial micro-muscle system integrated with Mott memory. The system utilizes the real-time regulation characteristics of the VO2 memristor on resistance to control torque and speed in real time (Figure 6). This artificial micro-muscle system makes possible future intelligent robots with the capabilities of feedback, response and adaptation.
(4) Polymer resistive memory. This type of memory gives full play to the flexibility and malleability of polymer materials and easy patterning and processing. It can seamlessly interface with electronic skin systems. It has unique advantages in the field of wearable devices, enabling wearable communication, processing and storage. Technology becomes a reality. In addition, flexible resistive memory has also emerged in the field of medical diagnosis. In 2014, "NatureNanotechnology" reported an electronic skin system integrating resistive memory devices and drug release devices (Figure 7). The system uses sensor elements to monitor the patient's physical signs in real time, saves, compares and feeds back the data to the drug release drive element through resistive storage elements, making the wearable diagnosis and health care system no longer just imagined in science fiction works.