ReRAM is emerging as the leading candidate to replace flash for many applications.
By Eran Briman, VP of Marketing and Business Development at Weebit Nano.
While flash memory is still the most popular non- volatile memory (NVM), several applications are beginning to adopt other types of embedded NVM technology, both because embedded flash can’t feasibly scale beyond 28nm and because of cost, power, and performance advantages.
Memories including Resistive RAM (ReRAM), Phase Change Memory (PCM), Magneto Resistive RAM (MRAM), and Ferroelectric RAM (FRAM), offer alternatives. Each technology has its own advantages and challenges in terms of cost, complexity, power and performance. However, ReRAM, offering the best balance, is emerging as the leading candidate to replace flash for many applications.
ReRAM technologies are already deployed in the market and new design starts are ramping up. Timelines for adoption vary by application and depend on the availability of ReRAM at different process nodes and foundries, as well as density and certification requirements. Let’s look at some applications where ReRAM is starting to gain traction.
Power management ICs
One of the first places ReRAM is gaining a foothold is in power management ICs (PMICs) and other high-voltage designs. Every electronic device has at least one PMIC managing system power distribution, and there is increasing demand for programmable PMICs to support trends like enhanced wireless charging and intelligent motor controllers. For these devices, PMICs must be smart and capable of running numerous algorithms, requiring a microcontroller (MCU) coupled with an NVM that is low-power, high-density, and cost effective.
To decrease cost and power, there is a move beyond two-chip solutions towards integrating the MCU with its NVM monolithically on a single die with power management circuitry. Embedded flash is too expensive, requiring up to 10 additional masks to manufacture. Flash is also difficult to integrate since it is manufactured at the Front-End-of-Line (FEOL), alongside the analog circuitry and influencing it, forcing companies to make compromises resulting in degraded performance, larger size, and higher cost. This is especially problematic for devices manufactured using Bipolar-CMOS- DMOS (BCD) processes, where significant attention is invested in optimizing FEOL power components.
ReRAM is a Back-End-of- Line (BEOL) technology, so it doesn’t interfere with the PMIC design. Unlike flash, which must be adapted to each variant, a BEOL technology like ReRAM can also be adopted once for a geometry and work with all the variants. And since ReRAM only requires two additional masks, it is significantly more cost-effective. PMICs, audio CODECs, and other high voltage designs with ReRAM are already in production.
IoT, MCUs and Edge AI
The market for low- power IoT devices using intelligent MCUs continues to grow. The Ericsson Mobility Report estimates that the number of connected IoT devices will more than double from ~15.7 billion in 2023 to ~38.8 billion in 2029.
Such ultra-low power devices must support increasingly sophisticated programming. They need NVM with the requisite ultra-low power consumption, performance, low total cost of ownership, and high data retention even in harsh or inaccessible conditions. When these devices must also perform simple edge AI tasks, it’s even more critical to reduce power consumption and costs.
Designers are increasingly integrating NVM into MCUs and System- on-Chip (SoC) devices at 28nm and below— from TinyML devices running simple edge AI in software, to devices with an AI accelerator, to full edge AI inference. Embedded ReRAM has significant advantages over external flash for all these applications and it’s beginning to sample now. Designers can reduce costs by eliminating external memory components and, with no need to fetch data from external memory, they can reduce power and increase system speed. A single die solution is also more secure against hacking.
the BEOL, there is no need for
compromise with power/analog components
In edge AI designs, in addition to code storage, ReRAM can store the synaptic weights needed for artificial neural network (ANN) calculations. Much of the power needed for ANN calculations is related to data movement between a system’s computing elements and memory modules. This data movement can be reduced by integrating dense, low-power NVM
like ReRAM closer to the computing elements (Near- Memory Compute). These arrays typically require significant on-chip memory, between 10Mb to 100Mb, depending on network size.
In the future, ReRAM will be a building block for neuromorphic (in-memory) compute, where computation occurs within the memory cell. As ReRAM cells have physical and functional similarities to synapses in the human brain, it is possible to emulate the brain’s behavior with ReRAM for fast real-time processing on massive amounts of data. This is orders of magnitude more power-efficient than today’s neural network simulations and will take more time to mature.
Automotive
The automotive market is extremely broad—ranging from tiny electronic control units (ECUs) for automatic windows to advanced autonomous driving technology. Over 1,000 chips control the various functions in a vehicle, almost all requiring some NVM.
Many automotive MCUs are moving to 28nm and below to run faster and handle larger amounts of data. They need NVM that supports fast boot, instant response, and frequent over-the-air (OTA) updates, often in harsh conditions. This NVM must have superior endurance, executing code quickly, reliably, securely, and cost- effectively while operating in extreme temperatures. ReRAM is the logical choice, and several major automotive chip vendors already deciding to rely on it. Due to qualification for specific temperatures and certifications like ISO26262 and AEC-Q100, these applications are still a couple of years out from mass production, but designers are already sampling automotive chipsets integrating ReRAM.
Secure applications
Whether it’s SoC security solutions such as physical unclonable functions (PUFs) and true random number generators (TRNGs), or embedded NVM for security applications like smart cards, ReRAM provides an inherently secure solution. Unlike floating gate devices like flash, ReRAM doesn’t use charge, so it’s more difficult to sense/change its internal state using electron beams. It is immune to electromagnetic fields, so ReRAM can easily withstand magnetic attacks. Since the ReRAM bit cell is deeply embedded between two metal layers integrated at BEOL, it is more immune to optical attacks. And because ReRAM can scale to small geometries, critical information can be embedded in the chip where there are no inter-chip communications to attack.
ReRAM is the future for many applications
Since nearly every electronic product requires NVM, the target applications for ReRAM are numerous. Embedded flash is reaching its limits, and ReRAM is emerging as the leading candidate for a new era of electronic devices. It’s already shipping in volume, and a growing number of foundries, IDMs, and semiconductor companies are committing to ReRAM as the leading embedded NVM in their future roadmaps.