Risky business: testing commercial nonvolatile memory chips for military applications
By Todd Wallinger
Commercial off-the-shelf (COTS) programs are all about saving money, and integrated circuits (ICs) provide COTS buyers with some of their biggest savings. While qualified military ICs can cost hundreds or even thousands of dollars, commercial versions often cost less than 10 dollars. The conclusion seems obvious: buy an off-the-shelf commercial IC, pay a third-party test house to screen it over the military temperature range, and then assemble it into a military system as though it were a qualified military device.
There’s just one problem. ICs are complex devices and a third-party test house may not know how to test them properly. This is especially true for NV (nonvolatile) memory chips.
NV memory chips are different from memories such as DRAMs or SRAMs in that they maintain their data after power is removed. They’re often used to store critical data, such as system configuration settings or error codes in military aircraft. Relying on third-party test houses to screen these ICs could result in devices that don’t meet the data sheet parameters or, worse, fail in the field.
An NVSRAM primer
One type of NV memory common in military applications is the NVSRAM (nonvolatile static random access memory), particularly if it’s based on SONOS (silicon oxide nitride oxide silicon) technology. That’s because these memories offer fast read/write speeds, long-term data storage, and excellent reliability—even at military temperatures. Alternative technologies, such as EEPROMs or battery-backed SRAMs, can’t withstand the higher temperatures demanded by today’s military applications.
An NVSRAM is simply an SRAM with a shadow NV cell adjacent to each cell of the SRAM. Short-term data is stored in the SRAM, which is why it can be written to and read from so quickly. These access speeds are comparable to those of standard SRAMs and are typically found in the 15-to-40-nanosecond range.
Long-term data, on the other hand, is stored in the NV. To get it there, the data must first be written to the SRAM. The entire array of data can then be transferred to the NV through one store operation. Because this involves injecting charges onto an insulating SONOS layer, it takes much longer than an SRAM write or read, often up to 20 milliseconds.
At this point, power can be completely removed from the device without losing the NV data. On power up, the NV data is transferred back to the SRAM through a recall operation. This typically takes around 50 microseconds. The data can then be read from the SRAM as normal.
The challenge of nonvolatility
It is this nonvolatility that makes testing NV memories so difficult. When a third-party test house screens an NVSRAM, they typically limit themselves to doing a write/read and store/recall at the temperature extremes.
Now this may be enough to ensure that the part works today, but it doesn’t ensure that the part will still be working 10 years from now. That’s because all NV memories lose some of their stored charges over time. Also, the storage layer can wear out with repeated use.
To screen NV memories, we have to consider two parameters unique to these devices: retention and endurance.
Retention
Retention is the amount of time a device can successfully retain data after power is removed. Depending on the NV memory technology, this can range from tens to hundreds of years.
Of course, no one wants to wait a hundred years to get the results from a simple test, so how can retention be measured? One way is to bake the part, then test it to see if the data has been retained, a fairly standard and well-accepted approach known as accelerated life testing. Baking a device for a specific amount of time is equivalent to holding the device at room temperature for a much longer time. The hotter the bake, the more time at room temperature the bake represents.
The problem is that the relationship between the bake conditions and the room temperature equivalent varies for each NV technology. Before going into production, the manufacturer will determine this relationship through hundreds of hours of testing and characterization. As a result, only the manufacturer will know what this relationship is. A third-party test house can certainly bake an NV memory, but it won’t know how hot or how long it needs to bake it to represent the lifetime of the device as specified in the data sheets.
Manufacturers of NV memories based on SONOS technology are able to take a couple of shortcuts to this process. Unlike EEPROMs, which tend to lose their stored charges catastrophically, SONOS memories leak a tiny but consistent amount of charges over time. By measuring this leakage, predictions can be made as to when all of the stored charges will be depleted, considerably shortening the test time. This, however, requires access to certain internal nodes of the IC. While the original manufacturer has access to these nodes, third-party test houses do not.
Another shortcut is to perform the store at a lower voltage. Weak cells will fail much sooner this way. But again, this requires access to the internal nodes of the device.
Endurance
Endurance is the number of times data can be stored in the memory cell. While the SRAM portion can be written to and read from an infinite number of times, the NV portion has a physical limit because the insulating ability of the storage layer eventually degrades. Again, with SONOS technologies, this degradation is gradual rather than catastrophic and can therefore be measured.
Endurance is usually specified in the hundreds of thousands to millions of store cycles. Unfortunately, the aging mechanism is nonlinear, so there are no shortcuts to measuring this parameter. To get a measurement of the endurance, the device must be taken to the end of its life.
For this reason, endurance testing is always done on a sample basis. Devices intended for military applications are taken from manufacturing lots that have already been characterized to have sufficient endurance over the full military temperature range. Commercial devices, on the other hand, may come from lots with lower endurance. Take one of these ICs to military temperatures and the endurance may be less than expected. Only the manufacturer knows what the endurance is for a particular manufacturing lot.
Trim
There’s one final complication that makes using commercial-grade NV memories in military applications a risky proposition. This is because there’s a tradeoff between retention and endurance. In order to find the sweet spot between them, many manufacturers trim—or individually program—each device.
Trimming is usually done by blowing on-chip fuses or programming dedicated transistors to adjust the resistance of an internal voltage divider. This sets the amplitude or duration of the electrical pulse used to store the data. Set the amplitude or duration too small and the device will inject too few charges into the storage layer, resulting in poor retention. Set them too high and the device will inject sufficient charges into the storage layer, but the layer will wear out much sooner, resulting in poor endurance.
Many manufacturers trim each device to optimize its lifetime for a particular temperature range. Use it at a different temperature range and its retention or endurance may suffer. Therefore, NV memories should only be used at the temperature range for which they’re trimmed.
An inexpensive solution
Used properly, NV memory chips are extremely reliable and are often the only type of memory that can maintain critical system data in military applications. But in these increasingly cost-conscious times, it can be difficult for military buyers to obtain them at an acceptable price.
The good news is that some manufacturers recognize this dilemma and have begun to offer an attractive solution. Simtek Corp. introduced a line of NVSRAMs identified as U grade. These devices have been trimmed and tested for functionality, retention, and endurance over the military temperature range, but have not been subjected to all of the environmental and electrical qualifications of a full military-qualified device.
So, with the skyrocketing prices of military-qualified ICs, it can be tempting to cut corners by buying commercial ICs and having a third-party test house screen them over the military temperature range. This practice introduces several risks, however, especially for complex devices such as NV memory chips. These risks include the possibility that the device may not meet the application’s long-term reliability requirements.
The solution is to find a manufacturer who offers an intermediate grade of their device or who can provide additional testing at a reasonable cost. This will give the user the best of both worlds: an inexpensive device and one which will meet the demanding requirements of today’s advanced military systems.
Todd Wallinger is a senior product engineer at Simtek Corp. in Colorado Springs, Colo. Contact the company online at www.simtek.com.
