Electronic Enthusiasts Network reported (text / Zhou Kaiyang) Fast, non-volatile, high endurance and low logic state switching power consumption, all these characteristics converged on a general-purpose memory was considered impossible in the past because of the physical characteristics are contradictory. Traditionally, fast, high-endurance memory is often paired with fragile and easily lost logic states, and even requires constant refresh, such as our DRAM dynamic random access memory. Non-volatile logic states often require a lot of power to complete the switching, which will slowly destroy the structure of the memory over time and reduce its durability, such as our Flash flash memory.
Even so, the pursuit of all-purpose general-purpose memory has not stopped, and many innovative technologies have emerged in recent years, such as resistive ReRAM, magnetic MRAM, or phase-change PCM. These memory technologies have also made amazing progress, and more or less have entered the stage of commercialization. For example, Samsung recently announced the first realization of MRAM-based in-memory computing, which is likely to be popularized in AI and wearable devices in the future. However, although these technologies solve the problem of volatility, the problems of logic state stability and switching energy consumption still cannot be improved.
And recently a technology called ULTRARAM entered our field of vision, and declared that it has the hope of becoming the next-generation general-purpose memory.
The ULTRARAM here is not the SRAM Ultraram in the Xilinx UltraScale+ FPGA, but a brand-new general-purpose memory alternative technology proposed by researchers at Lancaster University in the UK. ULTRARAM is also a non-volatile memory technology, which finally realizes fast and ultra-low energy storage of electrons in the floating gate through the resonant tunneling heterostructure of the triple barrier. Researchers at Lancaster University have completed the realization of the technology on a silicon substrate, which is also an important achievement towards mass production.
The heterostructure of ULTRARAM is mainly based on III-V compounds, using the quantum well of indium arsenide and the potential barrier of aluminum antimonide to create a triple barrier resonant tunneling (TBRT) structure. The 2.1 eV conduction band shift of aluminum antimonide provides a barrier comparable to the silicon dioxide dielectric in flash memory relative to indium arsenide that makes up the floating gate. Just apply a low voltage of around 2.5V, and the barrier becomes transparent to the electrons. Coupled with extremely small capacitors, the switching energy loss per unit area of ULTRARAM is reduced by 100 and 1000 times compared to DRAM and flash memory, respectively.
In addition to this, high durability is another great advantage of ULTRARAM. Through further testing of the device, researchers at Lancaster University have come up with promising results of up to 1,000 years of data retention and 10 million program/erase cycles, both of which are comparable to Flash. Said to be more than a hundredfold improvement.
In fact, as early as 2016, Europe invested 1 million euros in the research of antimonide in data storage applications, and carried out a three-year research with the participation of four major European universities, one of which is UK Lanka. University of Sturt. Now that ULTRARAM has finally made a huge breakthrough in becoming a general-purpose memory alternative technology, does it have a chance to successfully replace traditional memory and storage? We might as well look at other attempts made in the industry before making a conclusion.
Attempts at non-volatile memory
I believe that everyone has more or less heard of Intel's Optane persistent memory. With its extremely high performance, Optane has gained a number of enterprise users. Although the vision given by Optane is very good, from a factual point of view, in Intel's third quarter financial report last year, the Optane business lost $473 million in the first nine months of 2020, while the same layout of 3D XPoint Micron also stopped development in the first half of last year.
All this shows that Optane's market response is not good, mainly because of its cost and scalability. Compared with competing products such as NAND, Optane is really an expensive choice in terms of price/capacity. Intel would rather give up the NAND business and maintain the loss-making Optane. First, because of its large size, it can withstand such losses. Second, It is they who are optimistic about the potential of Optane in smart NICs and DPUs. But now that Intel has abandoned Optane's plan for desktop PCs, it can't be called "universal" memory.
In addition, it is the MRAM we mentioned above. In fact, Samsung's breakthrough in MRAM has more or less alleviated the problem of energy consumption of traditional storage media, but MRAM has not solved the problem of capacity in the final analysis, and cannot be used as a real meaning. general-purpose memory scheme. Take Everspin, which focuses on MRAM. Although their MRAM solution has achieved GB level, it is mainly aimed at markets such as IoT and automobiles that do not have excessive demand for capacity.
Speaking of which, everyone must know that the ultimate success of this type of non-volatile memory is not performance, but cost, mass production difficulty and scalability. ULTRARAM is still in the research stage. Whether it can break out of the crowd of alternative technologies in the end depends on whether there will be obstacles in mass production.
Original title: Will the next generation of general-purpose memory be ULTRARAM?
Article source: [WeChat Public Number: Electronic Enthusiasts Network] Welcome to add attention! Please indicate the source of the article reprint.
Reviewing Editor: Tang Zihong