New storage technologies introduction

Applications such as the Internet of Things (IoT), Artificial Intelligence (AI), 5G, and Industry 4.0 are driving explosive growth in the amount of information. All information must be collected at the edge and processed and transmitted, stored, and analyzed at multiple layers from the edge to the cloud.

With the huge demand for data storage and transmission, traditional memories such as DRAM, SRAM, and NAND Flash are gradually overloaded, and the process miniaturization of traditional memories is becoming more difficult, driving the semiconductor industry to develop new memories with higher storage performance, lower cost, and process miniaturization. Three types of memory stand out - MRAM, RRAM, and PCRAM.

Memory, as an important component of semiconductor components, accounts for up to 20% of semiconductor products. The main conflict in the memory industry today is between the growing demand for end-product performance and the technology that has yet to make a major breakthrough, specifically, the huge performance difference between memory and external memory that is causing the main bottleneck in the performance improvement of electronic products.

At the same time, we do not want to let Moore's Law growth slow down to limit the growth of computing in the era of artificial intelligence, do we provide a new script for semiconductor design and manufacturing. This strategic thinking underpins the next generation of high-capacity memory manufacturing systems introduced today for the Internet of Things and cloud computing.

MRAM (Magnetic RAM)

MRAM (Magnetic Random Memory) stores data by magnetic field polarization instead of electric charge. The memory cell consists of a free magnetic layer, a tunnel gate layer, and a fixed magnetic layer. The magnetic field polarization direction of the free magnetic layer can be changed, while the magnetic field direction of the fixed layer remains unchanged. When the magnetic field direction of the free layer is parallel to that of the fixed layer, the memory cell shows low resistance; otherwise, it shows high resistance, and by detecting the resistance of the memory cell, we can determine whether the stored data is 0 or 1.

MRAM includes many directions of research, such as microwave drive, thermal drive, etc. Traditional MRAM and STT-MRAM are the two most important categories, which are both based on magnetic tunnel junction structure, but the way to drive the free layer flip differs, with the former using magnetic field drive and the latter using spin polarization current drive.

MRAM

For conventional MRAMs, since a magnetic field cannot be introduced in the semiconductor device itself, a high current needs to be introduced to generate the magnetic field, thus requiring additional bypasses in the structure. As a result, this structure consumes more power and is also difficult to integrate at high density (usually only 20-30F2). If polarized current drive, i.e., STT-MRAM, is used, no additional bypass is required, and thus power consumption can be reduced and integration can be significantly increased.

MRAM characteristics

Non-volatile: The magnetic properties of ferromagnets do not disappear due to power failure, so MRAM features non-volatility.

The infinite number of reads and writes: The magnetism of ferromagnets not only does not disappear due to power failure, but can be considered almost never to disappear, so MRAM can be rewritten infinitely as well as DRAM.

Fast write speed and low power consumption: The write time of MRAM is as low as 2.3ns and the power consumption is extremely low, enabling instant on/off and extending the battery life of portable devices.

High integration with logic chips: MRAM cells can be easily embedded into logic circuit chips, requiring only one or two additional steps in the back-end metallization process, which requires a photolithographic mask version. In addition, MRAM cells can be completely fabricated in the metal layer of the chip, and even 2~3 layers of cells can be stacked, so it has the potential to construct large-scale memory arrays on logic circuits.

However, the biggest drawback of MRAM is the interference between memory cells. When programming the target bit, the free layer in the non-target bit can be easily misprogrammed, especially in the case of high density, the overlap of the magnetic field between adjacent cells will be more serious.

PCRAM (Phase Change RAM)

Another new type of memory is PCRAM (Phase Change Random Memory), which is also a sandwich structure with a phase change layer in the middle (same as the optical disc material, GST). A characteristic of this material is that it will transform between crystallized (low resistance state) and amorphous (high resistance state), using the difference in conductivity between the crystalline and amorphous states of the material after the transformation to store information in a process that can be divided into two main steps, SET and RESET.

Note: The difference in resistivity between the crystalline and amorphous states of a phase change material is several orders of magnitude, making it highly noise-tolerant enough to distinguish between the "0" and "1" states. The phase change materials currently used by various organizations are sulfide (represented by Intel) and synthetic materials (GST) containing germanium, antimony, and tellurium, such as Ge2Sb2Te5 (represented by STMicroelectronics).

internal structure of PCRAM

When the material is in the amorphous state, the temperature is raised above the recrystallization temperature but below the melting point temperature, and then slowly cooled (this process is the key factor governing the speed of PCM), the material transforms to the crystalline state (this step is called SET). At this time, the material has long-range atomic energy levels and high free electron density, so the resistivity is low.

When the material is in the crystalline state, the temperature is raised to a temperature slightly above the melting point and then quenched and cooled rapidly, the material transforms to the amorphous state (this step is called RESET). At this time, the material has a short atomic energy level and a low free electron density, so the resistivity is high.

PCRAM characteristics

Low latency, balanced read and write time: Compared with NAND flash, PCM does not need to erase previous code or data before writing updated code, so its speed has an advantage over NAND, and the read and write time is more balanced.

Long life: PCM read and write is non-destructive, so its write resistance far exceeds that of flash memory, and PCM is used to replace traditional mechanical hard drives with higher reliability.

Low power consumption: PCM has no mechanical rotation device, and no refresh current is required to save code or data, so PCM's power consumption is lower than HDD, NAND, and DRAM.

High density: Some PCMs use a non-transistor design, which can realize high-density storage.

Good irradiation resistance characteristics: PCM storage technology is independent of the charged particle state of the material, so it has a strong resistance to space radiation and can meet the needs of national defense and space.

However, the current problems of PCM are: when the phase change material in a device cell is in a high temperature melting state, thermal diffusion may cause adjacent device cells to also undergo a phase change, thus leading to errors in stored information. Currently, diodes as selector tubes is a major choice for high density PCMs. However, its preparation process can lead to the formation of parasitic triodes between adjacent diodes on the same word line. In turn, the crosstalk current of the parasitic triode affects the data stability. When a material undergoes a transition between amorphous and crystalline states, its volume changes, which in turn may lead to stripping of the phase transition material and the electrode material in contact with it, and device failure.

PRAM has now evolved into another area: Intel and Micron jointly launched 3D Xpoint technology in 2015. 3D Xpoint technology's memory cell is indeed PRAM, but it has found a suitable selection of tubes, i.e., a 1R1D structure instead of a 1R1T structure, which is completely different from Samsung's direction.

3D Xpoint technology has achieved a revolutionary breakthrough in the field of non-volatile memory. Although it is slightly slower than DRAM, its capacity is higher than DRAM and 1000 times faster than flash memory. However, there are obvious disadvantages: 3D Xpoint uses a stacked structure, which is currently generally a two-layer structure. The more layers are stacked, the more masks are required, and masks account for the largest share of cost in the entire IC manufacturing industry. Therefore, from the manufacturing point of view, it is very difficult to achieve a 3D stacked structure with tens of layers.

RRAM (Resistive RAM)

RRAM has been studied a little later compared to MRAM and PRAM. Although this phenomenon was reported as early as 1962, it did not attract much attention from academia and industry. It was not until 2000, when the University of Houston published an article in APL on the "discovery of electrical pulse-triggered reversible resistance transition effect in pon-magnetoresistive oxide thin-film devices", that Sharp bought the patent and started the development of RRAM in the industry, which has attracted research from academia and industry ever since. RRAM has also moved from the laboratory stage to the corporate R&D stage.

A typical RRAM consists of two metal electrodes sandwiching a thin dielectric layer, which acts as an ion transport and storage medium. RRAM looks similar to PRAM, except for the principle of the intermediate transformation layer. Phase transition is the transformation of material between crystalline and amorphous states, while resistance transition is the detection of high and low resistance states of a structure by forming and breaking thin filaments (filament, i.e., conductive pathways) in the material.

crossbar RRAM

The actual mechanism of action varies greatly depending on the material chosen, but the essence is that external stimuli (e.g., voltage) cause ionic motion and local structural changes in the storage medium, which in turn cause changes in resistance and this difference in resistance is used to store data. The most accepted RRAM mechanism is the conductive filament theory, based on the conductive filament device will not depend on the device area, so its potential for miniaturization is large.

RRAM characteristics

High speed: RRAM erase speed is determined by the pulse width of the trigger resistor transition, generally less than 100ns.

Durability: RRAM read and write and NAND is different, using a reversible non-damage mode, which can greatly improve its service life.

Multi-layer storage capability: Some RRAM materials also have a variety of resistance states, making it possible to store multi-layer data when a memory cell, thus increasing the storage density.

The memory matrix of RRAM can be divided into two types: passive matrix and active matrix. The memory cell of the passive matrix is connected by a resistive element and a nonlinear element (usually using a diode), the role of the latter is to make the resistive element get a suitable voltage division, so as to avoid the loss of the memory cell reading and writing information when the resistive element is in a low resistance state. The advantage of this method is that the design is relatively simple and the process is well miniaturized, but the use of a passive matrix will make the adjacent cells inevitably interfere with each other. Active cells are controlled by transistors to read and write and erase resistive components, although good isolation of adjacent cells interference, its design is more complex, and the device can be poorly miniaturized.

In terms of capacity, these three new types of memory, MRAM up to 4Gb, PRAM up to 8Gb, and RRAM up to 32Gb, are still very different from flash memory but don't forget that all three are more than 1000 times faster than flash memory in terms of read and write speed.

Conclusion

Emerging memory technologies have been around for decades and are now evolving into a critical period where performance is more important in more applications. These emerging memory markets are expected to generate $20 billion in combined revenue in 2029. On the other hand, future process miniaturization and increased economies of scale will drive down prices. Emerging memories will be embedded in ASICs, microcontrollers (MCUs), and even computing processors as standalone chips, making them more competitive than existing memory technologies.

Emerging memory covers a wide range of technologies, with a focus on MRAM, PCRAM, and RRAM, but it is still too early to say who will win. While the future of emerging memory technologies looks bright, they are still having a hard time penetrating some of the more entrenched technology markets. Even if the economics improve, it will be difficult for emerging memory to overturn the dominance of existing markets. If it can't win on cost, it doesn't mean much no matter who has more technical advantages over these entrenched technologies.

In short, the development of AI, 5G, IoT, and Industry 4.0 has led to explosive growth in the amount of information, and the new computing needs are driving the development of memory towards higher capacity, higher read/write times, faster read/write speeds and lower power consumption. The semiconductor industry is actively investing in the development of new memory, which is expected to replace the three mainstream memory products, DRAM, Flash, and SRAM, in the future.