New memory types are being developed that combine the performance and endurance of DRAM, and the non-volatility, high density and lower power consumption of Flash. Emerging memory technologies introduce different operating mechanisms that create new materials engineering challenges.
MRAM is becoming a leading candidate for IoT devices as it can offer better scalability than embedded Flash memory, higher density than static random-access memory (SRAM), and lower power consumption than both SRAM and Flash. SRAM is not ideal for IoT devices because it requires as many as six transistors per memory cell, and active leakage power can be high. MRAM promises several times more transistor density, enabling higher storage densities or smaller die sizes. Also, MRAM can be designed into the back-end interconnect layers of embedded system-on-chip products (SOCs). It can be used to store the SOC’s operating system and applications, eliminating the need for an embedded Flash chip for this purpose, thereby reducing total system chip count and cost.
MRAM is based on a pair of magnetic film layers, one with fixed polarity and one that can be flipped, separated by an insulating layer. The cell is programmed by flipping the movable magnetic polarity and read by sensing the resultant change in resistance. The device is fabricated by using PVD to deposit with extreme precision at least 30 different metal and insulating layers, each typically 1-30 Angstroms thick. Each layer must be precisely measured and controlled. Magnesium oxide film is the core of the magnetic tunnel junction, a critical layer that forms the barrier between the free layer and reference layer. It must be deposited with 0.1 Angstrom precision to achieve low area resistance and tunnel magnetoresistance.
Cloud computing requires the highest computing performance possible, and AI training requires enormous amounts of data to be brought close to machine learning accelerators, which accordingly have large on-chip SRAM caches supplemented by large, off-chip DRAM arrays—which require constant power. Power usage matters to cloud service providers because data is growing exponentially as AI proliferates, and grid power is limited and expensive. PCRAM and ReRAM are leading candidates for cloud computing architectures because they can offer lower power and cost than DRAM along with higher performance than solid state and hard disk drives. Like 3D NAND, PCRAM and ReRAM are arranged in 3D structures, and memory makers can steadily reduce the cost of storage by adding more layers with each product generation.
PCRAM cells work by switching, under influence of an external voltage, between amorphous and crystalline states, which have different resistivities. ReRAM technology comes in several forms and a broad range of materials can be used in creating it. Some types of ReRAM are embedded in a metal filament within an ionic bridge. Others are formed as oxygen vacancies produced within a base material. Bits are stored in the resistive materials, often metal oxide; programming is done by applying electric current to the resistive materials; reading is done by sensing different levels of resistance. PCRAM and ReRAM also offer the possibility of intermediate stages of programming and resistivity to allow multiple bits of data to be stored in each memory cell.
Beyond the horizon of these edge and cloud applications, research is progressing on in-memory computing. The frequent matrix multiplication operations of machine learning can conceivably be executed within a memory array. PCRAM and ReRAM are good candidates because both have the potential for multibit-per-cell storage. Currently ReRAM seems to be the most viable memory for this application. The multilevel cell architectures promise new levels of memory density that can allow much larger models to be designed and used. Bringing these new analog memories to fruition will require extensive development and engineering of new materials, and Applied Materials is actively pioneering some of the leading candidates.