Everspin Technologies has introduced a new MRAM-based memory architecture aimed at simplifying how embedded systems handle code storage and runtime data. The UNISYST MRAM family combines both functions into a single non-volatile memory, targeting applications where performance, endurance, and reliability are all constraints at once.
The announcement comes as traditional memory approaches—particularly NOR flash paired with SRAM or DRAM—are starting to show limitations in modern embedded designs.
Moving Beyond the Flash + RAM Model
Embedded systems have historically separated responsibilities: non-volatile flash for code storage and volatile memory for execution and data handling. That model introduces overhead—both in hardware complexity and in system behavior.
As workloads increase, especially with edge AI and software-defined functionality, the drawbacks are becoming harder to ignore. Larger codebases, frequent updates, and the need for faster boot times are pushing against the limits of conventional flash.
Everspin’s approach with UNISYST is to collapse those roles into a single memory layer. The architecture allows both executable code and runtime data to reside in the same non-volatile space, removing the need for multiple memory types and reducing system-level complexity.
MRAM as a Unified Memory Layer
Magnetoresistive RAM (MRAM) has been used for years in niche applications where endurance and persistence are critical. What’s different here is the push toward higher densities and broader use cases that traditionally relied on NOR flash.
UNISYST extends Everspin’s existing PERSYST MRAM platform and is designed to drop into existing system architectures with minimal changes. It uses a standard xSPI interface, supporting up to octal SPI at 200MHz, which makes it compatible with many current embedded designs.
Initial devices are expected to range from 128Mb to 2Gb, placing them directly in the density range where NOR flash is commonly used today.
Performance Without the Usual Tradeoffs
One of the main limitations of flash memory is write performance. Erase cycles, page boundaries, and limited endurance all introduce bottlenecks—especially in systems that require frequent updates.
UNISYST addresses this by delivering non-volatile storage with significantly higher write speeds and endurance:
-
Read bandwidth up to 400MB/s
-
Write bandwidth around 90MB/s
-
Write endurance up to 10× higher than typical NOR flash
-
No erase-before-write requirement
This changes how systems can handle updates. Instead of batching writes or managing wear leveling aggressively, designers can treat non-volatile memory more like working memory—while still retaining data across power cycles.
Designed for Harsh and Long-Life Environments
The UNISYST family is positioned for applications where memory failure or unpredictability isn’t acceptable. Devices are expected to meet AEC-Q100 Grade 1 requirements and support at least 10 years of data retention under extreme temperatures.
That combination makes it relevant for:
-
Automotive control and logging systems
-
Aerospace and defense platforms, including satellite systems
-
Industrial equipment with continuous operation and high write activity
-
Edge AI systems that require local model storage and frequent updates
In these environments, consistency matters as much as raw performance. MRAM’s ability to deliver deterministic behavior—without wear-related degradation—addresses a key weakness of flash-based designs.
Edge AI Is Driving the Shift
One of the clearer drivers behind unified memory is the shift toward running AI models directly on embedded systems.
Models are growing in size, often reaching tens or hundreds of megabytes. Storing them is one challenge, but updating them efficiently during development or deployment is another.
Traditional flash introduces friction here due to write latency and endurance limits. A unified MRAM approach allows faster iteration cycles and more flexible deployment strategies, especially in systems that need to adapt in the field.
Simplifying System Architecture
Beyond raw performance, the architectural implications are significant.
By combining code and data storage into a single device, UNISYST can:
-
Reduce component count
-
Eliminate separate flash and RAM interfaces
-
Simplify PCB layout and routing
-
Lower power management complexity
For engineers working on constrained systems, these changes can translate directly into reduced design time and improved reliability.
Availability and Next Steps
Engineering samples of UNISYST MRAM are expected in the fourth quarter of 2026, with additional densities planned beyond the initial range.
Everspin is positioning this as more than just a new product line—it’s an attempt to move MRAM into a broader role as a mainstream memory option for embedded systems, particularly as software-defined functionality continues to expand.