The semiconductor memory market is once facing drastic shortages, mainly due to a surge in AI demand, that is pulling capacity toward HBM, DRAM and NAND solutions. Device makers see lead times increase and prices increased sharply. To understand what is really happening in the market and where MRAM fits into the picture, we spoke with Sean Dougherty of Everspin Technologies.
Sean Dougherty is vice president of sales at Everspin Technologies, where he leads global revenue-generating activities across product sales, foundry services and strategic partnerships. He oversees Everspin’s worldwide sales and application engineering teams, helping customers design next-generation MRAM devices into demanding commercial, industrial, aerospace and other high-reliability applications.
Q: We’re seeing a lot of focus on memory shortages again, especially tied to AI demand. From your perspective, what’s actually happening in the market right now?
What we’re seeing today is different from the last supply crisis. During that period, many were suddenly line-down on many semiconductor products at the same time.
Today, most folks are able to get supply, but lead times are much longer and prices are significantly higher. In some cases, those price increases can make the end product difficult to justify economically.
AI is pulling a lot of attention and capacity toward HBM, DRAM and NAND. We don’t play directly in those markets, but the impact reaches adjacent markets. Customers are starting to look ahead to 2026 and 2027 and ask whether they’ll get the supply commitments they need. That’s really what’s driving the concern right now. It is also driving very significant price increases if they are getting the supply.
Q: Many suppliers are shifting capacity toward higher-margin memory. What does that mean for customers who still rely on NOR and other legacy non-volatile memory?
It creates uncertainty. These are long lifecycle products, so customers are not redesigning them every year. If a supplier starts moving capacity to something more profitable, those customers start to question whether their part is still a priority.
What we’re hearing is not that supply is gone today, but that people are not comfortable with where it’s going and the price impacts.
Q: Are you seeing more customers actively think about second sourcing now? What’s driving that conversation?
Yes. They’ve already been through one supply cycle, and now they’re seeing the market shift again because of AI.
If you don’t have a second source, or if that second source becomes less reliable, you’re exposed. So, customers are starting to look earlier and make sure they have options, rather than waiting until there’s a problem.
Q: Where does MRAM realistically fit as a second source today? Is this something that requires a full redesign, or can it be a more straightforward swap in some cases?
We’re not trying to replace everything. Where we fit today is in NOR-type applications. In many cases, our STT parts are pin- and protocol-compatible with existing NOR devices, so it’s not a full redesign just to evaluate us.
That doesn’t mean it’s automatic. Customers still have to qualify it, but it’s a practical path in the right applications.
Q: When customers start evaluating MRAM, what tends to stand out most compared to what they’re using today?
Write performance usually comes up first. NOR reads are fine, but writes can be slow and you have to manage how you use the memory. With MRAM, that changes. Then you get into endurance and retention. You’re not dealing with the same write limitations, which simplifies how people design systems that update frequently or need reliable data over time.
Q: Do you think this is a short-term cycle, or are we starting to see a longer-term shift in how people think about memory and supply risk?
There will always be cycles in memory, but the way customers are thinking about supply has changed.