MRAM high-speed memory technology is considered as DRAM memory successor

Technology Trends Have you ever considered the Windows progress bar rolling over after powering on and off, and then logging in and opening the desktop is a matter of course? The reason why the operating system needs to redo the memory initialization operation each time it is turned on is that the commonly used memory is the memory of dynamic random access technology (DRAM), such as SDRAM, DDR, and DDR II. Kind of memory. An important feature of the memory that uses DRAM technology is that they are volatile memory, which means that once the power is turned off, the data inside it will disappear. In other words, the existence of data in DRAM memory is actually maintained by relying on continuous power supply to refresh.

Therefore, every time the operating system is turned on, it is always necessary to write a series of data to be used by the system itself again. This is the work done by the operating system during the boot waiting time. For DRAM memory, if you want to exempt this process, the power for memory refresh cannot be broken. The so-called sleep (sleep), in fact, the computer continues to consume power, but only less than normal operation.

However, the Toshiba Group in Tampa, Florida, recently demonstrated to the public a new type of memory, Magnetoresistive Random Access Memory (MRAM), which will make this a thing of the past.

Magnetoresistive memory and DRAM memory use completely different principles. The way that DRAM memory represents "0" and "1" is to determine how much power is in the capacitor. It not only needs to maintain power, but also periodically charges the capacitor to ensure that the contents are not lost. The storage principle of the magnetoresistive memory does not use capacitors at all. It employs two nano-sized ferromagnetic magnets and uses a non-magnetic metal layer or insulating layer to sandwich a metal conductor structure at the interface. By changing the direction of the two ferromagnetic bodies, the magnetoresistance of the underlying conductor changes. As soon as the resistance becomes larger, the current passing through it becomes smaller, and vice versa.

Therefore, only one transistor can be used to judge the current value at the time of power-up, and two different states of the magnetic field direction of the ferromagnet can be judged to distinguish between "0" and "1". Since the ferromagnet's magnetism almost never disappears, the magnetoresistive memory can be rewritten almost infinitely. The ferromagnet's magnetism will not disappear due to power loss, so it is not as volatile as normal memory, but it can continue to maintain its content after power down.

Globalfoundries published a paper at the 2017 VLSI-TSA conference to explain how to solve the challenges facing eMRAM and make it more widely applicable to automotive MCU and SoC applications.

After several foundries publicly announced plans to start production of Magnetoresistive Random Access Memory (MRAM) by the end of this year and before 2018, a recent foundry described how MRAM can be used to dramatically increase data retention for embedded applications. Ability.

At the recent "2017 VLSI Technology, Systems and Applications Seminar (VLSI-TSA) held in Japan, Globalfoundries discussed the research paper that Everspin Technologies switched to 22nm with embedded MRAM (eMRAM). Process node progress.

According to Dave Eggleston, vice president of embedded memory at Globalfoundries, the key breakthrough highlighted in the article is that eMRAM can store data via reflow at 260 degrees Celsius, maintain 125 degrees Celsius for over 10 years, and have excellent read/write durability at 125 degrees Celsius. Ability. This will enable eMRAM to be used for general-purpose microcontrollers (MCUs) and automotive SoCs. He said: "The magnetic layer has always lacked thermal stability. Therefore, if the problem of data preservation is resolved, a broader market can be opened up."

Eggleston said that while MRAM demonstrated non-volatile, high-reliability, and manufacturability at previous technology nodes, it has begun to face challenges when scaled down to 2x nm nodes and the BEOL temperature of compatible embedded memory. As described herein, magnetic tunnel junction (MTJ) stacking and integration can be optimized at a 400 degree Celsius, 60 minute MTJ patterned thermal budget and is compatible with the CMOS BEOL process.

Eggleston said that the three major foundries have introduced products using this technology, and customers have chosen Global Development's Process Development Kit (PDK) for design. The major wafer equipment manufacturers started investing in this area a few years ago because they believe it has sufficient commercial potential, so the tool can be used for deposition and etching of MTJs. Eggleston said: "They have already invested in large-scale fabs like us and small companies like Everspin to invest in and develop products."

At the same time, MCU customers began to seriously study how to use MRAM to strengthen its architecture. Eggleston said: "They achieved faster write speeds and higher durability." This allowed them to start using embedded MRAM in applications that may have used static random access memory (SRAM) in the past. He also pointed out that the 2x nm node is the sweet spot of the technology in view of the simplicity of the circuit and the manufacturing cost.

Globalfoundries' MTJ Stacking and Consolidation Optimized at 400 Degrees Celsius, 60 Minutes of MTJ Patterned Thermal Budget, and Compatible with CMOS BEOL Processes

The market opportunities for eMRAM are not much different from other emerging and existing memory technologies: new mass markets include mobility, networking, data centers, Internet of Things (IoT), and automotive. Eggleston pointed out that for Globalfoundries, the Internet of Things and the automotive market are more important. "We once said that the two major markets are largely the same, but as a foundry, we have achieved greater growth momentum in the automotive industry."

Embedded flash memory (eFlash) has always been the commonly used embedded memory, but there are a variety of emerging memory options in response to market demand. In addition to eMRAM, there are phase change memory (PCM), embedded resistor RAM (eRRAM), carbon nanotubes (CNT), and ferroelectric field effect transistors (FeFET). Eggleston pointed out that no matter which choice, we must balance the data preservation, efficiency and speed. Both CNTs and FeFETs show potential for growth, but they are still not mature enough, while PCMs are suitable for specific applications and cannot be widely used in embedded applications.

Eggleston said: "MRAM and RRAM have similar functions, both of which are calibrated memories in the later stages and can therefore be more easily implemented in logic processes." Available process technologies include processes that require large chips, FD-SOI or FinFETs. He also stated that eFlash can be built into the chip, but it will be more challenging if it is to be built on various technologies.

Eggleston said RRAM stacking is simpler because less material is needed between the electrodes. He said: "And it does not require investment in equipment like MRAM. MRAM does require some investment in capital equipment due to the complexity of stacking." However, he pointed out that RRAM cannot provide data preservation, speed, and data required by the broader market. Durability and other capabilities.

Eggleston said that MRAM is better than RRAM because of its versatility, because its material composition can be adjusted between electrodes. "You can adjust it for better data retention or to support faster write speeds and durability." He added that this tunable capability allows Globalfoundries to Enough to use eFlash in the field, you can also adjust the speed, so that it can be used as non-volatile cache for server processor and storage controller.

Original title next-generation memory eMRAM ready to go