Ferroelectric RAM (FeRAM) is a fast, non-volatile memory that’s stuck in the same trap as Optane – and currently going nowhere.
There have been – and are – several fast, non-volatile memory technologies that could potentially fill a gap in the memory-storage hierarchy between NAND and NOR on the one hand and DRAM on the other. Phase-change memory-based Optane has been the most recent candidate and spectacularly collapsed as a mass-market technology because it could never get its production volume up to a level where the chips were affordable. Weebit Nano’s ReRAM is another candidate, but first we’ll take a look at Ferroelectric memory technology.
Ferroelectric RAM materials have a spontaneous and stable electrical polarization which can be reversed by applying an external electrical field. Binary data is stored in the bistable – up or down – polarization states. There are several ferro-electric RAM technologies:
- FeFET – Ferroelectric Field Effect Transistor
- FeCAP – Ferroelectric Capacitor
- FeRAM – Ferroelectric Random Access Memory
- NVDRAM using a ferroelectric capacitor
Substack writer Mark Lapidus of MDL Associates says a FeFET resembles a traditional – and volatile – metal-oxide-semiconductor field-effect transistor (MOSFET) with hafnium oxide (HfO2) ferroelectric material integrated in the gate stack of a logic transistor. This transforms it into non-volatile FeFET memory transistor.
The oxygen atoms on a molecule of HfO2 are positioned logically up or down when an external electrical field is applied to the material, and form a stable electric dipole with a c10ns switching speed and 10⁹ endurance cycles.
A FeCAP is a capacitor with its dielectric; barium titanate (BaTiO3) or lead zirconate titanate (PZT), having ferroelectric properties. Two semiconductor institutes, Imec and the Goergia Institute of Technology, are researching FeCAP ferroelectric memory.
FeRAM is based on FeCAP technology and Micron’s NVDRAM (Non-volatileDRAM) also uses FeCAP technology.
None of these have a valid chance of replacing either DRAM or NAND in mainstream PCs and servers. DRAM is fast and volatile and its capacity limits are getting bypassed by high-bandwidth memory (HBM) with 3D DRAM a possibility in the future. NAND is slower than DRAM and its 2D Planar NAND capacity limits arrive as manufacturing processes shrink below 20nm or so. But layered 3D NAND has arrived and is blowing capacity limits away with very high capacity chips.
There is no feasible way for ferroelectric memory developers to reach the massive production output needed to get their per-chip costs down to affordability – as Intel’s Optane experience demonstrated so vividly. Instead the developers need to look at more specialized markets where the expense of their chip scan be absorbed more readily.
That brings us to the embedded memory market, with a memory device embedded inside a microcontroller (MCU) chip. Semiconductor foundries manufacture the MCU, the memory and other devices in a single process – a manufacturing line. Typically NOR flash is used for this memory as it is non-volatile and faster than NAND. However MCU manufacturing processes are shrinking below 28nm to the 22nm area, and NOR becomes much more difficult and expensive to make at these smaller process sizes.
This is an opening for NOR replacement tech, and ferroelectric memories are candidates for that.
There are two ways to integrate a ferroelectric memory with a microcontroller (MCU) in a semiconductor manufacturing process: at the front or the end of the manufacturing process. They are called FEOL (Front End of Line) and BEOL (Back End of Line).
Weebit Nano provides a good example of how to bring a replacement memory technology to market. It has worked with Skywater, a CMOS semiconductor foundry, to make its ReRAM module available to Skywater customers. Weebit and Skywater qualified Weebit’s ReRAM for volume manufacturing with fully functional demonstration chips produced. Weebit, which raised $40 million in April last year is, according to CEO Coby Hannoch, “in advanced discussions with many leading fabs and integrated device manufacturers. Funds raised, combined with our strong balance sheet of approximately $31 million cash at the end of December 2022, ensure we are well placed to transfer and qualify our ReRAM technology in Tier-1 fabs and foundries to capitalize on the growing global need for better performing memory technology.”
DRAM and NAND fabricator Micron is a semiconductor manufacturer and could build NVDRAM products for sale – if the market for them is tempting enough. It would have to see that the capital involved would bring a higher return than investing in either NAND or DRAM. That is a formidably big ask for what is niche embedded memory market technology. But unless a large foundry operator like Micron adopts ferroelectric memory, then it’s down to a Weebit-style startup striding down the long road to develop and prove its product technology through partnerships with foundries like Skywater.
Until either of these happens, ferroelectric memory is stuck in technology institute research labs and unable to launch.