인피니언 HV GaN
반도체 AI 인더스트리 4.0 SDV 스마트 IoT 컴퓨터 통신 특수 가스 소재 및 장비 e4ds plus

Characteristics and prospects of the three next-generation memories that have entered the commercialization stage: MRAM, PCRAM, and ReRAM

기사입력2019.06.13 10:27

| MRAM, a key memory candidate for AIoT devices
| PCRAM, DRAM-based DIMM, SSD replacement
| ReRAM, Value in Storage Applications


Three new types of memory—MRAM (Magnetic RAM), PCRAM (Phase Change RAM), and ReRAM (Resistive RAM)—are nearing commercialization after decades of research, bringing a boost to the semiconductor and computing industries. New materials are needed to produce these three types of memory, and innovations in process technology and manufacturing are needed.
Characteristics and Prospects of the Next-Generation Three Memories: MRAM, PCRAM, and ReRAM

The ideal semiconductor memory should provide all of the following features: ▲fast read speed ▲fast write speed ▲random access ▲low cost ▲3D scalability ▲low power ▲non-volatility ▲high endurance ▲high temperature tolerance ▲multiple states for multi-bit storage. No memory currently provides all of these features.

Emerging memories such as MRAM, PCRAM, and ReRAM are leading candidates to complement mainstream memories. Let's take a look at the features of these three next-generation memories that promise improved performance, power and cost savings, and are set to replace mainstream memory.


MRAM, a leading memory candidate for IoT devices
MRAM is not only fast, non-volatile, and low-power (magnetic technology), but also cost-effective because the memory array can be embedded in the back-end interconnect layer with just three additional masks.

MRAM is slower than SRAM, but it is fast enough to be used as working memory for many embedded computing applications. It is also fast enough to meet level 3 cache requirements.

Several leading logic and foundry companies have announced early production of SoC designs using embedded MRAM.

MRAM is becoming a prime candidate for memory in IoT devices, especially those that need to support AI computation. Over time, MRAM will be able to meet the high-temperature requirements of automotive applications in the embedded market.

From a technology and manufacturing standpoint, the biggest challenge for MRAM is precisely depositing the many thin film stacks required to form the magnetic tunnel junction (MTJ), the fundamental programming element for representing digital data. Many metal and insulating layers must be deposited using physical vapor deposition (PVD) methods in a clean, high vacuum environment below atmospheric pressure.

The core magnesium oxide (MgO) thin film layer must be deposited with a precise crystal arrangement under strictly controlled conditions. Even a single atom's height change can affect performance and reliability.


PCRAM, a new storage class memory
PCRAM is based on 'phase change' materials, which are switched from a highly amorphous material arrangement to a crystalline arrangement by heat as a programming mechanism.

PCRAM has the characteristics of random access, lower cost than DRAM, 3D scalability, and non-volatility. As a result, major memory companies are evaluating PCRAM for applications such as non-volatile DIMMs to replace some DRAM-based dual in-line memory modules (DIMMs) and expensive solid state drives (SSDs).

PCRAM requires multilayer materials to be precisely deposited to form critical structures. PCRAM layers are not as thin as MRAM, but they are very sensitive to impurities. This requires PVD process technology that can use a variety of materials and prevents fine particles and impurities. Once the PCRAM stack is formed, individual memory cells are formed using plasma etching, and encapsulation protects the exposed phase-change material.

Scalability is the key to driving the PCRAM cost roadmap. 2D scaling is used to shrink the critical dimension (CD) to 20nm half-pitch, and 15-16nm design rules are expected to emerge soon. 3D scaling of PCRAM is even more promising. The initial design utilized a two-layer stack, but the technology roadmap suggests four-layer and even eight-layer stacks are possible.


ReRAM, targeting high-density applications
ReRAM technology comes in many forms. Some are embedded in metal filaments within an ion bridge, while others are created by oxygen holes generated within the base material. The information bits are stored in resistive materials, usually metal oxides. Programming is performed by applying current to the resistive material, and reading is performed by detecting different levels of resistance. ReRAM can also use a wide range of materials.

ReRAM has shown limitations in endurance so far, so efforts are needed to identify the cause of failure so that materials and manufacturing technologies can be improved to ensure reliability. If this problem is solved, ReRAM will be able to provide high density and low cost for storage applications.

Today’s mainstream memories, DRAM, flash, and SRAM, have matured over decades and are continuously evolving. However, these memories are becoming more difficult to scale in terms of performance, power, and cost, among other things.

Computing continues to expand. Many people imagine a world where billions of low-cost computers are embedded in every industrial and consumer product, forming the Internet of Things, and where data stored in public and private cloud data centers explodes.

New memory technologies will use structures and new materials that are created using precise thin film deposition, measurement, etching/removal, and encapsulation in a clean, uncontaminated environment. Even a small atomic change can make a huge difference in the processing of these next-generation memories.
이수민 기자
기사 전체보기