Comment The Way to Unified Memory

Heinz Arnold, Chefredakteur Markt&Technik
Heinz Arnold, Editor-at-Large Markt&Technik

Unified memory is considered the "holy grail" of storage technology. Has a start-up from Dresden found it now?

Users have been longing for a new storage technology: as fast as SRAMs, as dense and cheap as DRAMs and non-volatile as flash memory. Established storage companies and startups have been chasing this unified memory for decades. But whether MRAMs, FRAMs, RRAMs, PRAMs, or the sometimes more exotic technology - no one has yet made the big breakthrough.

However, the search for alternatives is becoming increasingly urgent, especially in the field of embedded memory. The conventional flash types have many disadvantages, in particular, they are slow, consume a lot of power, and no longer scale. At the same time, the pressure is increasing because a new market is emerging: AI. But if you want to bring AI into edge devices, the very last things you want are high costs and hunger for energy.

There is now a promising new approach - from the hopeful startup FMC (The Ferroelectric Company) from Dresden. The special thing: the company does not work with new exotic materials that would have to be integrated into existing process technology, such as Nantero. Although their carbon nanotube technology has so far attracted investments in the three-digit million range, the company, which was founded in 2001, has still not produced any specific products.

On the other hand, HfO2,which FMC relies on, is "simple". It uses the well-known ferroelectric effect, but it is already being used billions of times. The second differentiating feature of the technology: the transistor (FeFET) itself is the memory cell; the scaling problem does not arise; the arrays can be constructed simply and cost-effectively. And on this basis, the storage types that are so urgently needed in the AI environment, such as Ternary Content Addressable Memories (TCAMs), can be realized. FMC could not only successfully replace an existing technology, but could quickly grow into a newly emerging market. After all, Google had not beaten Microsoft in the field of operating systems.

This could be more threatening for FMC: many other companies are also working on storage based on HfO2, which can be concluded from publishing activities. So it is important for FMC to take advantage of the AI hour and turn the technical lead in the embedded market into products. We have every right to wait with tense anticipation, to see whether a startup from Dresden will be able to point the way to unified memory