RRAM: A New Approach to Embedded Memory
9 comments
The emergence of the Internet of Things (IoT) and the insatiable demand for smart devices in every aspect of life is driving a complete overhaul of traditional wisdom in the microcontroller and embedded memory markets.
As electronic devices become smarter, the software code becomes larger and needs to be processed faster to handle the communication protocols, authentication, message generation, and historical backlog. The reality is now dawning on our industry that current memory technology just can't deliver upon this new generation of code storage capacity and performance demands, with embedded software code increasing quickly from a few KiloBytes to several MegaBytes.
With analyst firms such as Web-Feet Research predicting that the embedded memory market for consumer electronics will reach over $2.88 billion by 2018, the time is now to figure out a solution to this problem. If traditional memory technologies can't meet the demand, then what can? And with Flash so ubiquitous in consumer electronics designs, is it even plausible to consider replacing the existing worn-out technology?
The memory dilemma
When designing a new microcontroller or system-on-chip, designers must consider embedded memory capacity, organization, and integration when making architectural decisions. However, the next wave of system architects will also have to re-invent the way embedded memory blocks are integrated with logic processing units.
If you look at current Flash technology, the storage of information is based on charge density. Losing a few electrons can cause severe reliability issues, and when scaling below 25nm, Flash technology suffers severe endurance, retention, and reliability degradation.
In an effort to overcome the endurance, retention, and reliability issues inherent in Flash technology, current data storage systems use a high-end memory controller chip that can manage complex architectures and algorithms, such as error correction codes and DSP algorithms, wear-leveling, and bad block management, data duplication, logic to physical mapping, garbage collection, and DRAM buffers. While these workarounds are needed to support Flash technology, these complex techniques are not easily transposed in embedded systems architecture. As a result, there is an urgent need for new embedded memory technology that can effectively serve as a Flash replacement.
Resistive RAM (RRAM) for embedded memory technology
Fortunately, there are emerging new memory technologies that are capable of overcoming the inherent limitations of embedded Flash technologies while also delivering cost effectiveness, performance, low-power consumption, and small footprint. One of the most promising candidates is known as resistive RAM (RRAM) technology.
A typical RRAM device consists of two metallic electrodes that sandwich a thin dielectric layer serving as the ion transport and storage medium. The exact mechanism differs significantly among the different materials being used, but the common link among all RRAM devices is an electric field or heat causes ionic movements and local structural changes in the storage medium, which in turn causes a measurable change of the device resistance.
One type of RRAM, developed by our team at Crossbar Inc., employs a commonly used amorphous film, such as amorphous silicon (a-Si), as the host material for filament formation. The conducting "filaments" created during resistance switching are composed of discrete metal particles instead of continuous metal plugs. These properties lead to a number of performance advantages that promise to eliminate the many issues facing Flash.
Unlike Flash, a-Si RRAM can successfully scale down to 5nm while providing much better performance. And unlike traditional RAM, a-Si RRAM is more cost effective due its smaller footprint. By blurring the lines between the memory types, RRAM can simplify the architecture and hierarchy of memory subsystems as we know them, providing cheaper, smaller, denser, and faster memory.
Based on CMOS, RRAM memory can be tightly integrated with specific logic functions, such as security encryption algorithms for smart cards or biometrics applications. This integration enables the high-performance, massive parallel processing systems used in big data analytics and is ushering in a new era of mixed memory and logic system architectures.
RRAM layers can also be applied on top of each other in a 3D architecture, enabling large memory size on a small footprint. With specific architecture techniques, RRAM can be layered on top of CMOS logic gates enabling ultra-dense logic plus memory system solutions.
RRAM technology provides the cost competitiveness needed for consumer electronics, home appliances, wireless sensors networks, and even disposable electronics. Its improved read and write performance and its increased energy efficiency also make RRAM a candidate for mobile devices and wearable electronics.
As RRAM becomes mainstream, we will see smarter system solutions powered by larger embedded memory and more efficient microcontrollers that can enable a new class of applications between connected devices. RRAM is the way of the future and is what will finally make the much-touted IoT world a reality.
