FRAM Turns 68

Photo of a memory chip next to a paper matchPerhaps the oldest nonvolatile semiconductor memory type is the ferroelectric memory, which recently celebrated its 68th birthday.  FRAM predates flash memory, EEPROM, and even UV-erasable EPROM.  It’s even older than mask ROM, which wasn’t invented until 1967!

As a matter of introduction to the technology, FRAM, or ferroelectric memory, is a read/write nonvolatile memory technology that performs significantly better than either NAND or NOR flash:

  • It reads and writes faster and at lower power
  • It has significantly better endurance (1014 cycles)
  • It doesn’t require an erase before a write cycle
  • Write cycle times are about as fast as reads
  • A single byte can be written to any address at any time
  • It’s insensitive to most X-rays and other radiation

FRAM’s main problem is that it costs hundreds of thousands of times as much as either NAND or NOR flash, making it impractical for most applications.

Why is it so wretchedly expensive?  There are a few reasons.  The biggest one is that the materials used to make it are not standard semiconductor materials and were pretty nasty things to introduce into a wafer  fabrication plant (a “Fab”).  FRAMs are  currently produced using either lead zirconium titanate (PZT) or strontium bismuth tantalate (SBT).  Both lead and bismuth atoms are very mobile: If they get loose in a chip they will move around, contaminating whichever part of the chip they finally come to rest.  Fabs that use such materials need to take extreme measures to ensure that the entire fab doesn’t get contaminated, and the PZT or SBT that is used in a chip must be isolated from the rest of the chip with a barrier layer, typically of platinum.  These materials are exotic and they are troublesome, so process developers avoid them whenever possible.

Because of this issue these technologies have not been ramped to high volume, and this is the most important reason that the price is so high. Without high volume they don’t benefit from the economies of scale as do mainstream memories, and this is an enormously important factor in memory costs: Things that ship in low volume cost a disproportionately high amount to make.  (This is a problem confronting Intel right now with its Optane, or 3D XPoint, technology.)  The problem of scale is explored in depth in the new report: Emerging Memories Find their Direction just published by Objective Analysis and Coughlin Associates.

But I digress.  FRAM technology has hit a milestone unique to all semiconductor memory types: It has turned 68 years old.  That’s because it was invented by Dudley Buck at MIT and published in his May 16, 1952 doctoral dissertation. Buck’s original devices used individual crystals for each bit.  Later on, in 1955, researchers at Bell Labs decided to try putting multiple bits onto a single crystal, producing the 256-bit chip shown in this post’s first photograph, which was on the cover of the September 1955 issue of the Bell Laboratories Record.  (The Memory Guy thanks Nokia Bell Labs for finding and sharing this photo for reprint in this post.)

I should note that although it’s called “Ferroelectric” the technology uses absolutely no iron.  It just exhibits a hysteresis, illustrated below, that resembles the hysteresis in magnetism, so the scientific community gave it that name.  Magnetic hysteresis causes a piece of iron to retain its magnetic polarity after an external field has been removed.  A similar mechanism in ferroelectrics results in the displacement (vertical axis) of an atom within a crystal when an external field (horizontal axis) is applied, and the atom stays where it was (A or B) after the field is removed.

Hysteresis curve

At the time it was invented FRAM showed promise as a computer memory since it could be made without stringing together magnetic ferrite cores, a difficult process that was done manually.  At that time all computer memories were made of these  cores.  Cores gave way to SRAM in the late 1960s and to DRAM in the 1970s since these technologies proved easier to produce than FRAM.

Another photograph of the 1955 Bell Telephone Laboratories 256-bit FRAM appears below.  This appeared in the June 1955 issue of Scientific American Magazine.  This device may have been the first monolithic memory ever made.

Die photo of an FRAM in Scientific America Magazine, June 1955

Although the photo appears to show an integrated circuit, it is no more than metal traces on the front & back of a wafer of SBT.  It has no transistors or resistors, so it’s not an integrated circuit, it’s just a bunch of SBT FRAM bits.

Now here’s the mind bender: The first integrated circuit wasn’t invented until three years later, 1n 1958, by Jack Kilby of Texas Instruments!  Kilby’s circuit had one transistor, three resistors, and a capacitor.  The Bell Labs chip had 256 bits.

Early pioneers in FRAM technology were Ramtron and Symetrix, both of Colorado Springs.  Today PZT and SBT FRAMs are made and sold by Ramtron (now a part of Infineon), TI, IBM, Fujitsu, and ROHM’s Lapis Semiconductor.  Samsung announced FRAM engineering samples in 1999 and Matsushita announced an FRAM-based microcontroller in 2003, but it is unclear whether either company shipped products for revenue.

Fujitsu has shipped an estimated 4 billion FRAM-based RFID chips over the life of the product.  Panasonic ships FRAM-based NFC RFID cards in high volume.  The two of these RFID applications make it very likely that FRAM has shipped more unit volume than all other emerging memory technologies combined, MRAM, ReRAM, or even PCM (including 3D XPoint)!

Despite the technology’s past challenges, FRAM is regaining popularity.  In 2011 researchers at the University of Dresden’s NaMLab found that Hafnium Oxide (HfO) could be made to exhibit ferroelectric properties.  HfO is used in all mainstream CMOS logic processes beginning with the 28nm node.  It’s a well-understood material that is already in very high volume production, so it already benefits from the economies of scale.  This opens up a whole new world of possibilities for ferroelectric memories!  Today a new surge of FRAM development has begun in a large number of research labs  both in academia and at leading semiconductor manufacturers.  NaMLab’s research has  spun out into a new company, Ferroelectric Memory Company, dedicated to HfO FRAM development and commercialization.

So a big “Happy Birthday!” to FRAM!  Who would have imagined that it would have predated all other semiconductor memory technologies, and even the integrated circuit itself?

For in-depth  information on FRAM and other emerging memory technologies please see the page for the new Objective Analysis emerging memory report: Emerging Memories Find their Direction where the report can be purchased for immediate download.

4 thoughts on “FRAM Turns 68”

  1. Great story, Jim, and beautifully documented. Thanks for this nice addition to NVM history. Brian Berg, Flash Memory Summit Technical Chair

Comments are closed.