Memory Startup Brings FeFET Solutions to NVM Market

Article By : Nitin Dahad

With hafnium oxide present in the gate insulator of every CMOS transistor, FMC's FeFET memory aims to turn any logic transistor into a memory cell...

The Ferroelectric Memory Company (FMC), a startup in Dresden, Germany, has raised $20 million in an oversubscribed series B funding round, to bring its ferroelectric field-effect transistor (FeFET) memory solution to the non-volatile memory market.

Ali-Pourkermarati-FMC
Ali Pourkermarati

In an interview with EE Times, Ali Pourkeramati, CEO of FMC, said there’s considerable interest in its FeFET memory. The company has exclusive license to two fundamental FeFET patents through the Technische Universität Dresden (TUD). The company was spun out from TU Dresden in 2016 by its co-founders Stefan Muller and Menno Mennenga.

“We started off looking to raise $5 million, but subscriptions kept coming,” Pourkeramati said. “We had to stop at $20 million. All the investors are strategic investors with an interest in memories, so are important to us,” Pourkeramati said. The funding round was led by new investors M Ventures (the corporate venturing arm of Merck) and imec.xpand, with participation from SK Hynix, Robert Bosch Venture Capital, and TEL Venture Capital. Existing investor eCapital also participated in this round.

The new investors will provide support to FMC throughout the semiconductor value chain to bring its ferroelectric memory technology to market. The company plans to expand its team in Dresden, as well as to start international expansion, including into the US and Asian markets.

Current memory technologies are not meeting some requirements for higher density, lower latency, and lower power consumption. Interest is growing in non-volatile memories (NVM), including ferroelectric memories, to fill those needs. Market research firm Yole Developpement expects the market for emerging NVM to reach US$ 6.2 billion by 2025. In its Emerging Non-Volatile Memory 2020 report, the analysis firm said that over $4 billion of this market will be from the standalone emerging NVM market, driven by two key segments, low latency storage and persistent memory; the embedded market segment would be over $2 billion.

Pourkeramati said, “The rise of artificial intelligence (AI), internet of things (IoT), big data, and 5G demands advanced next-generation memories that enable high speed and ultra-low power consumption, while being compatible with leading-edge CMOS logic processes guaranteeing reduced manufacturing costs.”

Ferroelectric memory has been around for years, but the biggest challenge has been scalability. FMC says it overcomes that challenge by exploiting the properties of crystalline hafnium oxide (HfO2), which in its in its amorphous form is already the gate insulator material of every CMOS transistor ranging from planar to FinFET. FMC’s patent-protected technology makes it simple to transform amorphous HfO2 into crystalline ferroelectric HfO2. This way, every standard CMOS transistor and capacitor can be turned into a non-volatile memory cell, a ferroelectric field-effect transistor (FeFET) or capacitor (FeCAP).

How FeFET works - FMC
The FeFET shows nonvolatile characteristics due to the fact that the two stable polarization states of the ferroelectric gate insulator modify the threshold voltage even when supply voltage is removed. Accordingly, the binary states are encoded in the threshold voltage of the transistor. (Image: FMC)

Pourkeramati explained the ferroelectric memory in the market to date has been based on PZT material. “Foundries didn’t like this because it is difficult to use, needs dedicated furnaces and machines. That makes PZT not scalable and costly.” By contrast, he said that HfO2 is already used in logic gates, with DRAM companies all using it. “Hence we can transform any logic transistor into a memory cell.”

He said that some years ago that Qimonda discovered that when annealed, HfO2 could change its polarization; the company patented it, and the patents were taken over by TU Dresden when Qimonda went bust in 2009. Eventually, FMC acquired the patents, and when Stefan Muller became interested in the technology that is when they decided to spin out FMC. Pourkeramati added, “FMC got the exclusive license of the two fundamental patents for FeFET applications, and in 2017, we licensed this to GlobalFoundries, with whom we worked to develop the technology.” He said the company is also working with other foundries.

FMC memory market entry
FMC plans to tackle the embedded memory market first, working through the foundries. Eventually it will then target the standalone market. (Image: FMC)

Pourkeramati said the company’s FeFET technology addresses both embedded and standalone memory for compute and storage, which he said is close to a $150 billion total available market in 2025. The company will first target the embedded market, with first product in 2023. “The market’s goal is to place the technology as close as possible to the CMOS logic node. With our FeFET, this comes as close as you can get. By distributing memory around the system on chip, you’re improving performance.”

“We have created IPs for all the different markets, including storage class memory. Our business model is to license our design and device IP to the OEMs and IDMs, as well as the process IP to the foundries.”

Further explaining its differentiation in the market, Pourkeramati said, “Fundamentally, we are enabling the scalability of the technology efficiently as you transfer from one technology to another. Also, many emerging technologies, such as MRAM, overlook endurance and data retention. We’ve already measured greater than 1011 cycles and data retention of 1000 hours at 125 degrees C; the expected limits of the material are 1015 cycles and 10 years at 175 degrees C [though these have not been tested yet].” He added that because its field driven the FeFET technology is low power (less than 1fJ per bit for both read and write), and switching speed is less than 1 ns. In addition, costs are lower because it only needs two additional lithography masks compared to say more than 10 masks for an eFlash memory cell. “Compared with, say MRAM, we don’t need additional memory elements and don’t need expensive tooling.”

HfO2 FeFET properties
Tested and projected properties of ferroelectric hafnium oxide (HfO2). (Image: FMC)

FMC has already made significant progress in the development of its non-volatile memory technology promising to offer superior performance compared with state-of-the-art and emerging memory solutions. It is currently working closely with major semiconductor companies, as well as with foundries in the US, Europe, and Asia.

In addition to its high speed, ultra-low power, CMOS logic compatibility, reduced manufacturing cost, and extreme temperature stability, FMC added that its technology provides complete magnetic immunity and high radiation resistance. FeFETs and FeCAPs can be integrated into CMOS production lines using existing equipment without the need for extra capital expenditures.

Leave a comment