Dresden memory startup to debut at Semicon Europa
The company is in the process of being spun out from the nano- and micro- laboratory (NaMLab) at the Technical University of Dresden. It is currently listed as The Ferroelectric Memory Company (FMC) although CEO Stefan Mueller told eeNews Europe said that the name may change during the formal company creation and registration process.
The company is the product of work at NaMLab on the ferroelectric effect in thin films of silicon-doped hafnium dioxide. That work was, in turn, based on a discovery made in research at now defunct DRAM manufacturer Qimonda in 2007 by Tim Boeske that hafnium dioxide, if prepared in the right way could be made to demonstrate a ferroelectric effect. Hafnium oxide is well known as an insulator material used for high-k metal-gate (HKMG) transistor structures.
FMC has been formed by NaMLab to commercialize the work and has taken over a publicly-funded program that will provide €500,000 (about $565,000) to cover development over the period April 2015 to September 2016. Meanwhile the small group of engineers that have formed the company are looking for early-stage investment and potential partners, Mueller said.
FMC is working on scaling up work that team members had done at NaMLab. "So far we’ve shown 100 bits that work now we want to make 1k to 1Mbyte arrays together with peripheral circuitry." This will be done on multiproject wafer runs in 28nm CMOS in 2016, he said. The group already has a close relationship with Globalfoundries Inc., which operates a wafer fab in Dresden.
Mueller said that advantages that hafnium-based ferroelectric memory shows over rival non-volatile memory technologies would be a lower energy switching per bit and lower cost. It could also show superior speed of switching. The technology shows an endurance cycling of 10^5 cycles in the array although there are hopes that this can be improved.
"The memory switching is not based on current flow but on field effect," said Mueller. It is also a bulk material effect, unlike some rival non-volatile memory technologies that are based on the making and breaking of filamentary conducting connections through an insulating layer. Such filament-based memories can have issues with thermal stability.
In addition, because ferroelectricity is in hafnium oxide, which is already being laid down as an insulator for transistors in CMOS it is already part of the manufacturing process flow. This compatibility with CMOS logic could give ferroelectric hafnium an easy route into providing embedded non-volatile memory for 28nm logic.
"The adder in terms of cost is just one or two non-critical masks," said Mueller. "We are aiming for the performance of embedded flash at the cost of a multi-time programmable memory," he added.
However, the physics of memory effects at the nanometer-scale is prone to uncertainty and hafnium dioxide is a transition metal-oxide so could some form of Mott transition or resistive RAM effect be responsible or partially responsible for results?
Mueller said that when research first started ferroelectricity was not known in hafnium oxide and that some academics had been skeptical. However, the experiments and analysis all supported the ferroelectric effect and that model predicted the behavior.
Related links and articles: