Embedded Processing Trends, Part 1: Memories that Last Forever? - Embedded.com

Embedded Processing Trends, Part 1: Memories that Last Forever?

The level of innovation in companies throughout the supply chain has increased significantly in the past few months. Engineers are now turning their attention away from marginal improvements that were designed to keep products coming out in the pipeline to more long-term achievements that can bring a new paradigm to the market.

Many end equipment manufacturers look to chip manufacturers for the basic technological leaps that will enable the next generation of products. This series of articles will look at three significant trends emerging in one of the most critical and competitive sockets in modern electronic systems, including: nonvolatilve memory, hyperintegration and new packaging approaches. First we'll look at non-volatile memory.

What is “non-volatile memory?
In short, it's memory that does not require power in order to retain stored data. In the past decade, the market for this capability has been limited to a few applications, mostly in automotive, smartcard, medical and space applications, and companies have been willing to pay a premium because the fabrication processes have not been efficient or cheap.

However, there are larger emerging trends like energy harvesting, mesh networked wireless sensors, building automation and security, product durability, and deeply embedded applications (where manual maintenance would be near impossible) driving many existing and new end equipments to require memory with low power, high endurance and radiation resistance.

With existing memory technologies in the embedded processing/SOC market, there are two persistent problems. First, the processor speed, efficiency and size have outpaced the available memory technology, thus forcing designers to implement complicated modules in the architecture and workarounds in the hardware. Second, many processors today run at a very low voltage (1-3V), yet

Flash-based memory (most common memory used today) needs more than 10V to write to memory. This pains engineers as they have to design in large charge pump architectures that are costly in terms of the space on the die, which in turn increases costs of the chip.

As shown in Figure 1, below , several different technologies have emerged, including, Phase Change Memory, Magnetoresistive RAM, Ferroelectric RAM and SONOS Flash, and the next few years will determine which will be successful and which will fail. Each type has its pros and cons, and they are all one small step away from an explosion of market acceptance in the embedded market.

Figure 1. Overview of Nonvolatile Memory Types

Phase Change memory
Phase change memory (PCM or PRAM) technology emerged from research done in the sixties, and most people have probably interacted with PCM thousands of times in a more familiar form like rewriteable CD-RWs or DVD-RWs.

The basic principle is an amorphous substance (chiefly made using an element in the Oxygen/Sulfur family on the periodic table known as chalcogenide glass) that transforms into an organized crystal structure when heated (through current injection for chip memory and lasers for optical memory).

When the materials change from amorphous to crystalline, the resistivity and reflectivity change drastically. Today, companies are developing chip memory using the resistivity change in the material. PCM or PRAM chips are available today (although cost prohibits all but the most necessary applications), but due to the temperature instability and cost, no one has developed an embedded PCM on a processor.

Benefits of PCM
1. Access times as fast as 100ns
2. Alterable at the bit level (RAM)
3. Write/erase speeds up to 30x Flash
3. Does not require separate erase step, but slower than SRAM in write speed (latency and bandwidth)
4. Scalable with new lithographies in foreseeable future
5. Up to four states for each cell which allows double the density (implemented in a lab only)
6. Up to 1 million write cycles
7. Data retentionup to 300 years at 85 degrees C

Drawbacks of PCM
High current density (>107 A/cm2) required to change memory state
Temperature sensitivity, especially for multi-bit storage type
Only programmable in system ” not factory programmable due to high flow temperatures erasing the memory
Long term resistance and threshold voltage drift (important for multi-bit storage)
Timeline to mass production (and cost effective) capability for an embedded SoC is still likely several years away

Magnetoresistive RAM (MRAM)
With this technology the storage is also not encoded electrically, but instead contained in a magnetic state. The bit of information is derived from the difference, or lack of, between a permanent magnet plate and an electromagnet whose polarity is adjusted through two write lines adjacent to the plate (or “cell”) creating a magnetic field that is picked up by the cell.

Memory is read by selecting a cell with a transistor and then measuring its electrical resistance (thanks to the magnetic tunnel effect). Right now, multiple companies are producing standalone MRAM chips today, and one has SoCs with MRAM on their roadmap. However, no company has a clear timeline for release of these chips.

Benefits of MRAM
1. Newer topologies can go up to 65nm process node
2. Densities approaching DRAM
3. Universal memory: No distinction between RAM section of memory and program memory ” can partition at any ratio
4. Access speeds up to 150 MHz at 90 nm
5. Theoretical access time down to 2ns
6. Resistant to radiation

Drawbacks of MRAM
1. Difficult to produce chips below 180nm due to complex write methods required (toggle method and spin-torque-transfer required for 65nm)
2. No foreseeable scaling beyond 65nm
Vulnerability to ambient magnetic fields
3. Expensive due to the complex architecture and precise spacing needed between write lines and plates
4. High current needed to generate the electric field necessary to write to the memory

Ferroelectric RAM
FRAM (also called FeRAM or F-RAM) is not actually related to iron in any way. The memory is stored as the state of a dipolar molecule (usually lead zirconate titanate or PZT film). There is a nonlinear relationship between the dipole state and voltage applied that resembles the hysteresis loop of ferromagnetic materials, and thus the name was born.

Figure 2. FRAM Memory Cell

Today, only two companies license the technology from Ramtron and are producing FRAM in volume, one of which is Texas Instruments. The highest density FRAM chip available is a 4 MB device, produced by TI for Ramtron, but there are also several integrated devices for smart card applications that are beign produced by Ramtron and the other licensee. So far, the largest chips only have 16kB of embedded FRAM, but the high volume production capability has been proven.

Benefits of FRAM
1. Low Power ” can be programmed down to 1.5V with no charge pump
2. Reliability: intrinsic anti-tear (data not corrupted when power lost in the middle of an operation)
3. Durable: Up to 100 trillion write cycles
4. Faster (by 1,000 time) than conventional Flash memory (50ns per word)
5. Small cell (1T-1C) = small chip size
6. Soft Error Rate is nearly zero means it is very radiation tolerant
7. Compatible with existing digital CMOS processes and easily ports to new process nodes
8. Flexibility: “universal memory” can be used as cache, data or code memory because the memory is high speed and non-volatile
9. Security: No charge pump for light flash perturbation, high speed of write/read resists scans and monitors
10. Density: Higher density than Flash (6T), but lower than DRAM (1T-1C)

Drawbacks of FRAM
1. Smaller sizes (<32kb embedded="" today)="">
2. Destructive read operation means every read is also a write (mitigated by the extremely high number of write cycles in life of memory)

SONOS Flash
The differences between SONOS Flash and regular Flash are that it uses an insulating layer of silicon nitride (Si3N4) instead of a polysilicon layer, and it places a salicide “cap” on the source and drain of the MOSFET. The name SONOS (silicon-oxide-nitride-oxide-silicon) refers to the layering of the transistor.

The ONO layer traps the charges and allows them to remain stored on the chip for an extended period of time without leaking away, thereby making the memory non-volatile. This process only adds a few masks and steps to the normal CMOS fabrication process (salicide on the source and drains, ONO layer being the most notable). In addition, the ONO layer combats the leakage current problem that stops current Flash technology from functioning beyond 45nm. There are a few companies currently manufacturing the technology today.

Benefits of SONOS Flash
1. Very similar to CMOS FET production so little change is needed for high volume production (<10 added="" steps)="">
2. Scalable beyond 45nm node
3. Up to 1011 write cycles
4. Security: Cannot read memory via electron microscope or ion scans
5. Retention up to 15 years at 85 degrees C or 100K write cycles

Drawbacks of SONOS Flash
1. Same high voltage required to program as existing Flash, still requires the complex circuitry to be added on to silicon die
2. Write time up to 5-10 ms and up to 10 ms erase time ” requires complex architectures to integrate with higher speed cores which increases cost of silicon
3) Same security issues through the charge

It is obvious that one of these could be the next Flash. There are just one or two simple factors blocking mass market acceptance. This could be as little as a marketing push, finding the right emerging equipment to support the R&D costs, or it could be a seemingly minor leap in technology.

Right now, from looking at the pros versus the cons and seeing what is already in production today, it looks like FRAM has pulled into the early lead. Nonvolatile FRAM memory has a lot of potential not only to fill the needs in new applications like building automation, but it can also offer many advantages in addition to a reduction in costs for existing applications.

The next two articles in this series focus on two other emerging embedded processor trends: (1) “hyper-integration,” which involves the astonishing amount of analog and digital components being added onto modern SOCs, and, (2) the emergence of innovative and incredible new package types that enable deeply embedded control and electronics to be virtually invisible.

Jacob Borgeson is currently a product marketing engineer for microcontrollers at Texas Instruments. He has a special interest in integrating low power functionality with wireless operation in many standard applications; especially biomedical devices. He works with Universities and TI engineers to help enable the best educational environments and next generation products through directed research.

References:
1) Cypress SONOS Technology by Krishnaswamy Ramkumar
2) Technical Advantages of Nonvolatile F-RAM Memory in System Design, Ramtron Corporation
3) The Basics of Phase Change Memory Technology, Numonyx, Inc,
4) MRAM Wikipedia page
5) Past, Present, and Future of MRAM by Shehzaad Kaka, NIST Magnetic Technology
6) FRAM – New Generations of Non-Volatile Memory, Texas Instruments.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.