FRAM - true unified memory

Somewhere in the beginning of 2007 I won in a lottery held by EPN magazine an evaluation board from Ramtron sporting their most powerful 8051 clone, the VRS51L3074, which could be programmed and debugged with a small USB piggyback board. The very special thing of the Ramtron controller was the 8 KB of non-volatile ferroelectric memory (F-RAM) it included.
Instead of putting the board in a box, I actually used it in a project. I used the F-RAM to store some parameters in. The evaluation board also had several other F-RAM chips on it, but I did not use those. Once the project finished (it was published in the January 2009 issue of Circuit Cellar) I (almost) forgot about F-RAM and never heard from Ramtron again.

This year at the Embedded World show in Nuremberg I again came across F-RAM, now spelled FRAM, at the Fujitsu stand. It turned out that they started licensing the technology from Ramtron and that they had put some of it in their MB95R family of 8-bit MCUs. Fujitsu also offers stand-alone FRAM devices up to 4 Mbit.

This week Texas Instruments announced their brand-new MSP430FR57xx family that includes up to 16 KB of FRAM, licensed at the same source. The people at TI are very excited about it and have high expectations for the future.

So what exactly is FRAM and why would you be excited about it?

The F in FRAM stands for ferroelectric (and not ferromagnetic nor Fujitsu) because it “uses ferroelectric film as a capacitor for storing data. Possessing characteristics of both ROM and RAM devices, FRAM features high speed access, high endurance in write mode, low power consumption, non-volatility, and excellent tamper resistance.” (Source: Fujitsu)
How it all works at the atomic level is not so important here, let’s jump to the bits interesting for people wanting to use FRAM in a design.

First there is speed. Compared to flash memory FRAM is way faster, TI claims a 100 times faster. According to Fujitsu, FRAM is 30,000 times faster than EEPROM.
Secondly, there is power consumption. FRAM uses very little power, 250 (TI) to 400 (Fujitsu) times less than flash or EEPROM. This is active power. Interestingly enough FRAM has a slightly higher leakage current than flash memory, so power consumption in sleep mode is slightly worse.
Less energy and higher speed means cost savings, not only on the power budget, but also on the production budget for the high-volume people that measure the time it takes to program a device on the production line.

Then there is endurance. Flash & EEPROM memory cells are specified for some 10,000 write cycles, whereas a FRAM memory cell can be rewritten more than a trillion times! (Fujitsu is a bit conservative here, TI rather optimistic.) This kind of endurance combined with the higher speed means that FRAM can be used in place of SRAM.

FRAM works with low programming voltages down to 1.5 V. Replacing a block of flash memory by a block of FRAM of the same density then saves die space because FRAM doesn’t need a charge pump. Smaller means cheaper. Also, because of the low programming voltage new applications become possible where the higher flash programming voltages are currently considered dangerous.

Finally, FRAM is less sensitive to external fields and alpha particles than flash memory because the memory cell capacitor’s dielectric is much thinner than the one of a flash memory cell. The thinner the dielectric the less distance there is to develop a dV over and the smaller are the chances to (accidentally) modify a bit.

All these advantages were known from the start, so why did it take so long before FRAM made its way into non-Ramtron MCUs? Well, I don’t know about Fujitsu, but according to TI the main reason was the incompatibility of the technology. Five years ago FRAM was done in 350 nm technology, which was too big to use it in TI’s chips. Now they brought it down to 130 nm, which is a perfect fit.


This is a picture of a demo board showing off the low-power capabilities of the new MSP430 family, targeted at energy harvesting applications. Battery backed SRAM, flahs or EEPROM can all be replaced by FRAM, the speed and low power requirements of FRAM make external super caps unnecessary. Higher speed equates to shorter up times so ultra-super-low-power applications seem to be ideal.

Since anyone can license this technology from Ramtron, we can expect other parts from other manufacturers to appear in the next months or years.

Jeez, why didn’t I buy Ramtron stock at the time when nobody cared?

New kit on the blog

At the last Embedded World show it was Cortex-M3 everywhere and I discovered that Fujitsu (now?) also sells cheapish development kits for their Cortex-M3 products. I obtained one of these kits, the SK-FM3-100PMC, to check out. The kit contains an SK-FM3-100PMC board, an AB type USB cable, a CDROM and a sheet with additional information. It does not contain the promised CD/DVD with Softune Workbench mentioned in the additional information.

On the board you’ll find an MB9BF506N processor (100 pins, 80 MHz, 512 KB flash, 64 KB RAM, 8 multifunction serial ports, 2 CAN ports & USB host & device, plus of course the usual timers and PWM-like peripherals, DMA and what not). Available connectors are UART-A RS-232 (9-pin female subd), UART-B USB (type A), CAN (9-pin male subd), USB host, USB device (type A), 20-pin JTAG (as usual the wrong way around for my Keil pod, Keil even gets this wrong on their own boards. Actually the cable is upside down on the pod) & Trace and headers for all the processor’s pins. On-board gadgets are two 7-segment LED displays (I always like that) and two square white push buttons.

The board can be powered from a multitude of sources: externally feeding a 1117 5V regulator, JTAG and both USB-A connectors. The processor itself will work from 2.7V up to 5.5V, meaning that this is one of the rare Cortex-M3 devices that can run directly from 5V.



Lets go!
Insert the CDROM and connect the board to the PC using the UART-B connector. A pretty well written Readme file pops up that guides you through the process of setting up a development environment for the kit. Driver installation is very easy, Windows can do it all alone (except for the usual “Do you really want to do this?” nagging). When the board is powered the 7-segment display starts counting up to 99 and then stops. Pressing the INT0 push-button makes it count down to 00, pressing INT1 makes it count up again.

The readme also describes how everything should be packed in the box, which is cool if you want to put it away properly.

The board has a reset LED, interesting. It lights up when you press the (square blue) reset button. Such an LED is maybe helpful to detect situations where the board is kept in reset without you wanting to?

The CDROM contains a free serial port terminal program SKwizard. This utility only knows about four COM ports, when my board is on COM20… Luckily the flash programming tool FM3 can handle COM ports up to nr. 20 (phew…). Yes, I know you can assign a different number to a virtual COM port, but this is poor/old school programming. The other flash programming tool (there are two) allows port numbers up to 255.

To test ease of (re)programming the board I first tried the FM3 flash tool. This one uses the serial port for programming and is pretty straightforward to use. The only weird thing (besides the COM port numbers) is that the file picker wants an S-record file, but only knows the extensions MHX, CNV & AHX, and not the SREC files that come with the kit.

The second tool for direct USB programming over the USB device port is a bit more difficult to get started, because it needs a driver that is not on the CDROM and I could not find it on the Fujitsu website either. Contacting Fujitsu taught me that the driver is hidden in the installation folder of the Flash USB Direct tool (on my PC: c:\Program Files\Fujitsu\FUJITSU USB DIRECT Programmer\driver\). I was also told that no drivers are available yet for Windows 7, but they are coming. Once the driver installed (COM 22!), I could reprogram the board. To my surprise direct USB programming turned out to be as slow (or fast, but I prefer slow in this case) as serial port programming. However, I did like the friendly message after successful programming: “It ended normally completely”. (Probably to be pronounced with a Japanese accent.)

Example code is provided for IAR and Keil compilers. I did have Keil uVision 3 with a ULINK2 pod, but you need version 4 to open the project files, so I downloaded & installed an upgrade of 312 MB. Now with my brand new version 4 I could indeed open the example projects.

What the people at Fujitsu forgot to mention is that you have to copy the examples to your hard disk otherwise uVision will not be able to build them as it tries to create files in the project folder. More annoying is that they also did not mention that you have to manually select the device - for some reason the project does not include the device – otherwise you cannot flash the board after a build and you receive confusing error messages. So, before building a project, select the device: Project -> Options for Target, click on the Device tab and select the MB9BF506N. This will also load the correct flash algorithm file. Now build the project and load it in the processor with your ULINK2, ME or PRO pod.

If you look closely at the project files you will see a file named FLASH_DEBUG_README.txt. This file says that you should copy an FLM file from the project to the Keil environment, but it seems that uVision 4 already contains usable flash files that get loaded when you select the target device.

I could not try trace recording, because you need a ULINK Pro for that.


The board in a Keil uVision 4 / ULINK2 environment.

Now that I am finally ready to start developing for this board it is a shame to discover that there doesn’t seem to be a nice Fujitsu library encapsulating all the peripherals of the chip and the board; some sort of a BSP as people like to call it nowadays. The best they have come up with is an empty template project, a bit meager.

Documentation includes full schematics of the board, but it is lacking a mechanical drawing with the positions of the break-out connectors that seem to be placed on a convenient 0.1 inch grid. So measure carefully when you design an extension card for it.