Ferroelectric random access memory (FRAM) is widely known as a non-volatile, stand-alone memory technology that has been a part of the semiconductor industry for more than a decade.
In recent years, integrated circuit manufacturers have been considering FRAM as a strong contender for embedded, non-volatile memory, as an alternative to flash technology. This article discusses key technology attributes of FRAM while exploring specific use cases that demonstrate FRAM’s advantages.Today there are multiple memory technologies that have the potential to change the landscape of embedded processing. However, none so far have surfaced as a strong contender for replacing flash technology in microcontrollers (MCUs) until FRAM.
What is FRAM?
FRAM is non-volatile memory that has power, endurance and read/write speeds similar to commonly used static RAM (SRAM). Information stored in an FRAM cell corresponds to the state of polarization of a ferroelectric crystal that can hold its contents even after the power source is removed. This is what makes FRAM truly non-volatile. Also, since the energy required to polarize a crystal is relatively low when compared to programming a flash cell, FRAM writes are inherently lower power than flash.
Figure 1. FRAM allows for continuous ultra-low-power data logging and supports more than 150,000 years of continuous data logging (vs. less than 7 minutes with flash)
Here are a few typical applications that use microcontrollers with flash technology today. Let’s look at how leveraging FRAM-based MCUs, rather than flash-based MCUs, bring cost, energy and efficiency optimization.
Data logging:
Here are a few typical applications that use microcontrollers with flash technology today. Let’s look at how leveraging FRAM-based MCUs, rather than flash-based MCUs, bring cost, energy and efficiency optimization.
Data logging:
A typical data logger application such as a temperature data logger can sample at rates anywhere between 1-1,000Hz. Now consider the write time of a single byte in flash memory is approximately 75µs.
In comparison, FRAM technology can be written to at a rate of about one byte every 125 nanoseconds. This is close to 1000 times faster than flash! Now consider the application reaches the end of a flash segment and needs to move to the next one, suddenly there is a 20 millisecond latency while waiting for a segment erase to complete.
The erase latency does not apply to FRAM as it is not required to pre-erase FRAM bytes between writes. A 20 millisecond latency every segment does not seem prohibitive until we calculate how significantly it impacts the maximum write speed. For the purpose of this discussion, consider that the block of memory being written to is 512 bytes in length. A flash memory block can be written 26 times per second including the time taken to complete an erase cycle every time 512 bytes are written. This brings us to a total speed of 13kBps [1].
In comparison, a 512 byte FRAM block can be written to at speeds greater than 8MBps [2]. Not every application requires such high write speeds, but consider if your target application was required to write only 1kB every second, the MCU with flash technology would spend 7% of the time staying active to perform the write. However an FRAM MCU would complete that task in 0.01% of the time allowing the MCU to remain in standby 99.9% of the time, providing significant power savings.
Energy harvesting
Many applications today are focused on using cleaner and greener energy, energy that is derived from natural sources such as sunlight, vibrations, heat or mechanical change. Such applications rely on small bursts of energy that provide power in short time intervals and the MCU is usually down to the wire in terms of how many lines of code can be executed before power is lost. Flash-based applications pay a premium in power, not only because of higher average power while accessing flash, but also because of higher peak power during flash write events.
This peak power is mainly due to the usage of a charge pump and can reach values of up to 7mA, making non-volatile writes virtually taboo in the energy harvesting world [1]. With FRAM there is no charge pump; therefore, no high current writes. The average power when writing to FRAM is the same as when reading from or executing out of FRAM (i.e. there is no penalty for non-volatile writes, making FRAM a truly flexible option for energy harvesting applications .)
Figure 2. All-in-one: FRAM microcontrollers deliver maximum read, write, power and memory benefits
RFID Tags:
Radio frequency identification (RFID) tags are making an appearance in many places: store shelves for displaying prices, name badges at conferences and on industrial automation floors to mark and identify objects on a conveyor belt. Some of these applications require memory writes up to 100 times a day.
Consider a byte of flash memory with a typical endurance of 10K write/erase cycles. To achieve 100K write/erase cycle endurance, the application will have to set aside 10 bytes of flash memory for every one byte of data, meeting the endurance requirements at the cost of high redundancy.
In comparison, an FRAM memory byte can endure 1015 write/erase cycles – 100 billion times more than a flash byte [3]. For applications that require high endurance in the order of millions of write/erase cycles, FRAM’s endurance specification is unmatched by other embedded non-volatile memory technologies available today.
Handheld metering:
Blood glucose metering is one example where loss of power is highly critical. In the case of power failure due to a depleted battery, the meter is required to save a time stamp, store the readings at the time of failure and perhaps even perform a few math functions before shutting down.
Consider a flash-based metering application with a battery that is depleted of charge, the power drop can be approximated to about 300mV in 0.01 seconds. In this time, up to 80K FRAM bytes can be written compared to about 8K bytes in flash. However this is without factoring in the high peak and average current requirements of a flash write, which will drain the battery rapidly, bringing down the backup capability significantly.
Another use case of system backup in power fail events is in energy metering where the energy reading needs to be preserved in non-volatile memory until power is restored. In such cases, the power usage during system backup is critical as backup battery sources are expected to last up to 10 years.
The list of applications where FRAM not only provides differentiation, but may also be the only viable option, is as diverse as is vast. To test drive an FRAM-based MCU check out the MSP430FR57xx series from Texas Instruments Incorporated (TI). Samples can be obtained for free and the MSP-EXP430FR5739 FRAM experimenter’s board is available online for $29.
FRAM can lower system cost, increase system efficiency and reduce complexity while being significantly lower power than flash. If your existing flash-based MCU application has energy, write speed, endurance or power fail backup constraints it may be time to make the switch to FRAM.
References:
