The demand for automatic system data storage is increasing in embedded applications. This trend is driven both by systems manufacturers' needs to have higher traceability, as well as any possible information for diagnostics, failure analysis and application debugging, and also from end users' desire to store application parameters needed to optimize application usage, to trace relevant historical series, and for security reasons.
As a result, the concept of “Ubiquitous Computing,” the pervasive diffusion of digital logic in any object of our life, couples more and more with “Ubiquitous Memory.”
In many countries, local regulations drive the storage of sensible process parameters into a non-volatile (NV) and safe support for quality or security critical areas, like transportation, data and voice carriers or food and drug supply chains.
We typically refer to these kinds of applications as “data logging,” including both stand-alone data recorders (devices for which the acquisition and the storing of data is the unique function) and more common data logging applications embedded in more complex electronic systems, supporting many other functions (a PC for instance).
Matching data logging apps to memory architecture
Dedicated NV memory support is clearly mandatory for stand-alone data logging devices, but separate memory support for data logging is sometimes present in complex systems that already have a huge amount of NV memory on board, like hard disk drives (HDD) or NAND Flash, for their main functionalities.
The reasons for this architectural choice vary, depending on the specific application, but are mainly related to security, reliability and independent accessibility (in case of main system default). The right selection of NV memory support can be the critical success factor for data logging applications.
Flight Data Recorder. The “Black Box,” originally designed for traceability, maintenance and failure analysis, has gained popularity in other transportation areas outside the aeronautic sector, such as automotive (e.g., Event Data Recorder) and others with regulated safety standards.
Network Control . Increased need for security in both physical and virtual spaces, such as access control applications, have amplified demand for massive data logging in secure memory support.
Pay-Per-Use Technology . Businesses that run on a pay-per-use model, (e.g., printers or fax machines) need cost-effective, secure data logging support.
Wireless Sensor Networks . Typically, wireless networks employ on-board NV memory to store the latest data and download measured data into a central control unit, enabling the data download in batch mode and ensuring the possibility for local data reading without the central system connection. Industrial applications for this kind of sensors are getting widely diffused in the market.
Consumer Electronics . In the consumer electronic and PC segments, applications that embed a traceability system storing the tear level parameters – sometimes even the number of power-on cycles performed – are getting widely diffused. Gaming systems (console, arcade and gambling) have data logging requirements as well, and many GPS systems feature a Flash memory data logging module on board.
Choosing a memory technology
The requirements for NV support in storing information for different applications varies greatly depending on the data logging characteristics of the specific application: type and dimension of the data, protocols, frequency of recording, necessity for replacement or incremental growth of the log file, reading and writing timing performance, security features (in some applications the log data must be un-erasable), data retention requirements, integration in the same memory of other NV storage needs of the application like code storage.
The first key parameter driving the selection of the right memory media is, of course, the density.
The required density depends on the type of data, sampling frequency and total period for which the data must be stored. Also the usage model and the algorithm of recording is relevant; some applications require a permanent log of all the progressive data, others instead can download the data continuously or in batch mode and then replace it with new one.
If the application requires a large amount of data (in the order of many Gb), the recommended solution must be NAND Flash memory or even HDD. This is the typical case for Black Box like applications requiring the storage of large amounts of parameters and in some cases even multimedia content (e.g., voice, pictures, etc.).
The advantages of Solid Sate solutions versus electromechanical are related to the mechanical robustness, due to the lack of moving parts, and for lower access time. In fact, the main drawback is related to minor program erase cycles endurance (wear leveling is mandatory) and data retention performance. In choosing the right NAND device a trade off between cost, density and performances must be taken into account while selecting a Single Level Cell (SLC) versus a Multi Level Cell solution (SLC).
For very low density (up to few Mb) both an EEPROM or a NOR Flash device can be used. EEPROM has poor cell scalability and is therefore limited to low densities, and a lower read speed, but it offers bit alterability and better endurance. Page Erasable Serial Peripheral Interface (SPI) Flash memory featuring the Byte Write command (EEPROM emulation) is recommended for applications that need to migrate from EEPROM to Flash for higher density and improved reading/programming speed requirements.
In the range from 10 Mb up to few Gb, NOR Flash is generally the preferred solution. Also, EPROM is used in applications not requiring the onboard erase of programmed data.
Battery-backed Random Access Memory (RAM) is also used in niche applications requiring very high performance in read and write timing, but in most cases the RAM solution is overcome by Flash mainly for power efficiency and reliability reasons.
NOR Flash solutions are available in the market in two macro-families: Parallel interface and Serial interface. SPI Flash has many advantages for data logging applications, despite the fact that it's currently limited in density to the 128Mb threshold due to the traditional addressing protocol composed by three bytes.
The key parameters for memory solutions in the density range up to few Gb are listed in Table 1 below , for data-logging applications, as previously mentioned, the most important are Read Data Throughput, Programming Time, Endurance and Data Retention.
|Table 1. Typical Key Performance Parameters for 3 V commercial devices (public datasheets)|
Why SPI Flash
The main advantages of SPI solutions are the low pin number, which simplifies the board routing (or bonding for multi-chip solutions), decreases the total system cost and enables very compact solutions, and the memory size scalability — the same protocol (and pinout) can be used for different densities and even for different devices on the same bus.
The typical disadvantage of a traditional SPI Flash solution, with respect to Parallel Flash, is the poor random access time (even if the latest generation of SPI Flash is addressing this topic through various tricks like Multi I/O Reading and XiP modes), but in typical data logging applications the data have a “store-and-download” usage model, no code execution is required on Flash, and random access time is not an issue.
For these reasons, SPI Flash memory can fit many of the data logging requirements, representing a cost effective, scalable and easy to design solution.
The right choice of the SPI Flash must take into account many variables; we will briefly address the main ones.
Density. A wide range of SPI density is available on the market, ranging from half Mb up to 128Mb. And once the above-128Mb addressing issue will be solved with a standard solution, many higher densities will come soon.
Supply voltage. Beside the embedded traditional Vcc range of 3V, many suppliers are starting to deliver 1.8V SPI Flash, which is mandatory for wireless applications (also requiring a special focus on power efficiency).
Erase granularity. Sector (256 KByte, 64KByte or 32KByte), subsectors (4KByte) or page (256Byte) erasable SPI Flash are available. Of course, we must ensure that finer granularity implies bigger die size, and then increased cost of the memory. Higher granularity is recommended for data logging applications requiring a frequent replacement of small data.
Speed. Latest devices reach operating frequencies higher than 100MHz. This, combined with multi I/O reading, enables more than 400Mb/s of data throughput for continuous reading (typical in the store-and-download usage model). The random access time can be reduced in the newest generation, skipping the instruction overhead (XiP mode), but as previously mentioned, this is generally not a killer feature for data logging applications.
For programming and erase timing, typical values for floating gate technologies are less than 1 ms to program a 256Bytes page and less than 1s to erase a 64K Bytes Sector.
Protections. In some applications the log data must be un-erasable to prevent accidental or malicious modifications; for these applications, SPI flash can be found in the market with a one-time programmable protection feature. A unique ID feature is also available to prevent the original device replacement.
Packages. A broad SPI Flash package portfolio is available in the market, including very compact packages, thanks to the low pin number (e.g., MLP 8 2x3mm).
Endurance. Typical values are 100K cycles (program/erase) per sector. Optimized usage of programming and erasing algorithms enable long-term applications; the key point is to fill a complete sector with data before erasing it and distributing the cycling in all the sectors. Data retention specs are typically in the range of 10 to 20 years.
Integration. SPI Flash is easily integrated both in PCB and in system-on-package solutions due to their low pin count. Specifically for multi-chip solutions, Numonyx provides “Known Good Die” parts even with automotive quality standards.
Temperature range. Industrial (-40C to 85C) or automotive (-40C to 125C) temperature range can be selected.
Long-term support, together with the reliability and quality specs (few ppm) of the best suppliers of SPI Flash, are also very important for data logging applications.
FeRAM, MRAM and PCM are very attractive technologies for data logging applications. In particular, PCM, thanks to its merge of EEPROM advantages, NOR performances and a good scalability path, on top of bit alterability, enhanced speed (with respect to Flash in both Write and Read mode) and better endurance, is the best candidate in data logging applications for NOR replacement.
|Table 2. New nonvolatile technologies benchmarks|
The most promising next generation NVM technologies (Table 2, above ) today are 1) ferroelectric memories (FeRAM), which exploit the permanent polarization of a ferroelectric material; 2) magneto-resistive memories (MRAM), which use the electrical resistance change in a magnetic tunnel junction to indicate the memory state; and, 3) phase change memories (PCM), based on an electro-thermally induced phase-change transition in a chalcogenide alloy.
Their development has faced many challenges. Additionally, FeRAM and conventional MRAM, show scaling limitations requiring the introduction of alternative approaches (such as Spin-torque MRAM).
On the other hand PCM, thanks to the intrinsic good scalability that allows a process roadmap forecast down to the 2x nanometer note, can be considered one of the best candidates for next-decade NVM mainstream technology.
1) Non-volatile memory technologies: emerging concepts and new materials – Roberto Bez and Agostino Pirovano, Materials Science in Semiconductor Processing, Volume 7, Issues 4-6, 2004, Pages 349-355
Stefano Andreoli has been working in the technical marketing team of Numonyx, with a special focus in SPI Flash Memory, since 2007. Previously, Stefano worked in the R&D department of STMicroelectronics, committed to Flash memory components, compact modeling, and accenture technology solutions.