Thursday , 13 June 2024
Home Electronics: Technology, News & Trends Emerging Storage Technologies: Overcoming Limitations and Meeting Modern Demands

Emerging Storage Technologies: Overcoming Limitations and Meeting Modern Demands

Data storage

With the rapid succession of emerging applications such as 5G, artificial intelligence (AI), and smart cars, the market has placed higher demands on data storage in terms of speed, power consumption, capacity, reliability, and more. As a result, storage technology is constantly facing new challenges.

Among the current mainstream storage technologies, although DRAM is fast, it has high power consumption, low capacity, high cost, and cannot retain data in the event of a power failure, limiting its use cases. NAND Flash, on the other hand, has lower read/write speeds, and its storage density is significantly constrained by the manufacturing process.

To overcome the limitations of traditional storage technologies like DRAM and NAND Flash, barriers to storage technology are continuously being broken, and new storage technologies are entering the mainstream.

What Are the New Types of Storage?

Currently, emerging storage technologies aim to integrate the switch speed of SRAM, the high-density characteristics of DRAM, and the non-volatile properties of Flash. New storage technologies can be mainly divided into Phase Change Memory (PCM), Magnetoresistive Random-Access Memory (MRAM), Resistive Random-Access Memory (RRAM/ReRAM), and Ferroelectric RAM (FRAM/FeRAM).

PCM achieves different resistance values by changing the phase state of phase-change materials through thermal energy conversion, making it suitable for large-capacity independent storage applications. MRAM changes resistance by altering the direction of the magnetic poles in magnetic materials, making it suitable for small-capacity, high-speed, and low-power embedded applications.

RRAM utilizes the generation or closure of conductive channels in resistive materials to achieve resistance changes. Currently, it is mainly used for Physical Unclonable Function (PUF) and may play a role in future fields such as artificial intelligence and storage-computing integration. In recent years, storage-computing integration has gradually become one of the hot trends to address current storage challenges.

The aforementioned new storage technologies share some common characteristics, such as non-volatility or persistence, and all mainstream non-volatile memories originate from Read-Only Memory (ROM) technology. Some technologies can reduce costs by shrinking the size through the manufacturing process. They do not require the block erase/page write methods needed by Flash, significantly reducing write power consumption while improving write speed.

Phase Change Memory (PCM)

Picture of PCM

PCM achieves resistance changes by transforming the phase state of phase-change materials (such as the change from liquid to solid at 0°C). It is considered by some as a type of Resistive Random-Access Memory (RRAM). PCM features long life, low power consumption, high density, and good radiation resistance. Before writing and updating codes, PCM does not need to erase previous codes or data, resulting in improved read/write speed compared to NAND Flash. PCM is considered the most compatible and mature storage technology with CMOS processes.

For PCM, temperature, cost, yield, and other factors are key conditions for its technical breakthroughs. Additionally, PCM’s multi-layered structure allows phase-change materials to be compatible with CMOS processes, but this also leads to low storage density, making it unable to replace NAND Flash in terms of capacity. PCM technology is represented by 3D Xpoint, a joint development of Intel and Micron.

In 2006, Intel collaborated with Samsung to produce the first commercially available PCM chip. In 2015, Intel and Micron jointly developed a revolutionary storage chip called 3D Xpoint, with the former branding it as Optane and the latter as QuantX. 3D Xpoint technology is a non-volatile storage technology that differs from NAND Flash, which stores data by charging and discharging crystals through transistors. Instead, 3D Xpoint uses PCM phase-change materials to store data.

Intel and Micron claim that 3D Xpoint, while slower than DRAM, has 1000 times the performance of Flash, 1000 times the reliability, and 10 times the capacity density.

Unfortunately, with the closure of Intel’s Optane business, 3D XPoint technology is also consigned to history. According to the Tom’s Hardware report in November of last year, Intel quietly released the Optane SSD DC P5810X solid-state drive, which may be the last storage device based on 3D XPoint flash memory.

However, the industry is still developing PCM technology. In early last year, the Institute of Information Storage Materials and Devices (ISMD) at Huazhong University of Science and Technology, in collaboration with the Materials Innovation Design Center (CAID) at Xi’an Jiaotong University, developed a mesh-like amorphous structure PCM, with a power consumption of less than 0.05pJ, one-thousandth of the mainstream products.

Magnetoresistive Random-Access Memory (MRAM)

Picture of MRAM

MRAM is a technology based on the tunnel magnetoresistance effect. MRAM products are mainly suitable for special applications with low capacity requirements and emerging IoT embedded storage fields. This technology features unlimited read/write cycles, fast write speeds (as low as 2.3ns), low power consumption, and high integration with logic chips.

The mainstream MRAM technology is mainly represented by STT-MRAM (Spin-Transfer Torque Magnetic Random-Access Memory) introduced by the American company Everspin. Everspin is a company that designs, manufactures, and commercially sells discrete and embedded Magnetoresistive RAM (MRAM) and Spin-Transfer Torque MRAM (STT-MRAM).

In 2019, Everspin, in collaboration with the foundry GlobalFoundries, conducted trial production of 28nm 1Gb STT-MRAM products. In March 2020, they announced the extension of the joint development of STT-MRAM devices to the 12nm FinFET platform, aiming to further reduce the cost of 1Gb chips through process miniaturization. Everspin has deployed over 120 million MRAM and STT-MRAM products in data centers, cloud storage, energy, industrial, automotive, and transportation markets.

STT-MRAM uses the “giant magnetoresistance effect” in the tunnel layer to read bit cells. When the magnetic directions on both sides of this layer are the same, the resistance is low, and when the magnetic directions are opposite, the resistance becomes high. Compared to other emerging storage technologies, STT-MRAM exhibits excellent durability, extremely fast storage speed, and is considered one of the top-tier cache memories.

Another MRAM technology is SOT-MRAM (Spin-Orbit Torque Magnetic Random-Access Memory), which employs a three-terminal MTJ (Magnetic Tunnel Junction) structure, separating the read and write paths to provide higher durability.

Due to their characteristics of storage speed and durability, both of these memories are expected to be optimal choices in high-performance computing systems, such as data centers. However, for STT-MRAM or SOT-MRAM to be used as high-density memory, further improvements are needed in terms of memory cost and density.

Resistive Random-Access Memory (RRAM/ReRAM)

Picture of ReRAM

ReRAM is a non-volatile memory based on the reversible transformation of the resistance of non-conductive materials under the action of an applied electric field. This technology features high speed (generally less than 100ns), strong durability, and multi-bit storage capacity.

ReRAM is divided into various technical categories, including Oxygen Vacancy Memories (OxRAM), Conductive Bridge Memories (CBRAM), Metal Ion Memories (MeRAM), Memristors, and Carbon Nano-tubes (CaRAM). Representative companies include Crossbar in the United States, Panasonic, and Xinyuan Semiconductor.

Due to the random nature of conductive channels in the storage medium, ensuring uniformity in large-scale arrays for binary storage is challenging. Therefore, ReRAM is widely believed to meet the requirements of energy consumption, performance, and storage density for applications such as neuromorphic computing and edge computing. It is expected to be widely used in AIoT, smart cars, data centers, AI computing, and considered one of the best choices for realizing storage-computing integration.

Among emerging storage technologies, ReRAM technology is more suitable for adopting multi-level storage in storage units, contributing to reduced energy consumption and improved cost-effectiveness in storage computing. In recent years, international manufacturers such as TSMC, Crossbar, Intel, Fujitsu, Samsung, UMC, and Adesto have focused on this technology.

In December of last year, Infineon announced that the next-generation Aurix microcontroller would use embedded non-volatile memory, specifically Resistive Random-Access Memory (RRAM), instead of embedded Flash memory (eFlash). Samples of the Aurix TC4x series microcontrollers based on TSMC’s 28nm eFlash technology have been delivered to major customers, and the first batch of samples based on TSMC’s 28nm RRAM technology will be provided to customers by the end of 2023.

It is worth mentioning that Xinyuan Semiconductor, a domestic new storage enterprise, announced in June of last year that its “Xin-Shanwen” secure storage series products based on ReRAM had achieved commercial use in core components of industrial automation control. This marks the further industrialization of ReRAM new storage technology at the 28nm process node in China. In addition, ZTE Microelectronics and Rambus jointly established a joint venture, Hefei Ruikewei, for the commercialization of RRAM technology, with no news of mass production at present.

Ferroelectric Random-Access Memory (FRAM/FeRAM)

Picture of FRAM

FRAM technology utilizes the characteristic of ferroelectric crystal materials to achieve information storage with a hysteresis loop in the voltage-current relationship. Ferroelectric materials can be used for both capacitors and data storage in the gate oxide layer of CMOS integrated circuits.

FRAM technology has the characteristics of fast read/write speeds, long lifespan, low power consumption, and high reliability. With these features, it is becoming one of the future directions for storage development. FRAM products have been commercially validated in the semiconductor market and applied successfully in automobiles. Companies representing FRAM include Ramtron and Symetrix, Infineon, and Fujitsu Semiconductor.

Previously, researchers discovered ferroelectricity in hafnium oxide (HfO2), which was widely used in CMOS gate oxide layers. Because of its advantages, such as fast speed, data non-volatility, and ease of integration into CMOS, ferroelectric materials are being extensively researched as a candidate material for a new type of memory.

For decades, researchers have dedicated themselves to developing new storage technologies that can replace traditional memory. While the current new storage market primarily focuses on low-latency storage and persistent memory, lacking the ability to replace DRAM/NAND Flash, these new storage technologies, with their outstanding characteristics of superior performance, long lifespan, reliability, and high-temperature resistance, are expected to become a new choice in the field of memory in the era of explosive data growth.

1 Comment

  • We would like to thank you again for the beautiful ideas you offered
    Jeremy when preparing her post-graduate research and also, most importantly, with regard to providing each of the ideas within a blog post.
    Provided we had known of your site a year ago, we may have been saved the needless measures we were participating in. Thank you very much.

    toys for adults

Leave a Reply

Your email address will not be published. Required fields are marked *

Related Articles

China semiconductor industry

Mainland China’s Semiconductor Capacity to Surpass South Korea and Taiwan, Becoming the World’s Largest

According to Latest reports, the construction of new wafer fabs worldwide has...

Musk nueralink

Musk’s Brain-Machine Company’s First Human Test Subject’s Brain Device Malfunctions

The path to Musk’s “symbiosis of man and machine” is not without...

International Energy Agency

IEA: Batteries Are Changing the Game for Achieving Climate and Energy Goals

While renewable energy (generation) itself has become cheaper than coal and gas-fired...

Brain computer interface

Zhejiang University Makes Mind-Controlled Chinese Character Writing a Reality

On April 23, Zhejiang University’s Brain-Machine Interface Team unveiled their latest research...