In my last blog this past month, Single Event Effects (SEEs) with High Speed ADCs: Single Event Latch-up (SEL), we began our discussion of SEEs with high speed ADCs by taking a look at SEL which is latch-up induced by radiation. As we learned SEL tests are typically performed first since latch-up can be destructive. Latch-up mitigation techniques can be employed to work around destructive SEL but these countermeasures come at an increase in system cost and complexity. Typically it is desired to avoid a device that exhibits SEL especially if the SEL is destructive. Now let’s move to our next SEE topic which is single event transients (SETs).

If you recall from my blog back in December 2017, A Quick Overview of Radiation Effects, SETs are soft errors that are temporary in nature where the device recovers on its own without requiring a power cycle or device reset. Unlike SEL testing it is not required in the test standards to maintain a high case temperature for the device when evaluating SETs. Generally, the test is performed at ambient temperature conditions while the temperature is monitored and recorded. Just as the case with SEL testing, the SET evaluation is typically performed up to an energy level of 80 MeV-cm^{2}/mg and out to a fluence of 10^{7} ions/cm^{2}. In some cases, the energy level can be ≥ 120 MeV-cm^{2}/mg for the SET evaluation. It depends on the desired application and situation. The evaluation would still be completed out to a fluence of 10^{7} ions/cm^{2}. Once again we will take a look at the AD9246S 14-bit 125 MSPS ADC. As a reminder I’ve included the setup diagram again here for SEE testing.

The goal of SET testing is to determine the energy level where there is onset of SETs with the ADC as well as to find the energy level that is the saturation point for SET events. It is important to understand both the magnitude and the length of the transient events. The SET behavior of the device is monitored while exposing the ADC to increasing levels of radiation and recording the number, magnitude, and length of SETs.

In the case of the AD9246S there were four different ions used which provided five different energy levels. Below is a table detailing the ions, the ion angle, and the energy levels used to test the AD9246S. Notice that Xenon is used twice with two different angles. Using the angle allows for the effective LET to be increased without needing to change the ion. It is a pretty simple calculation to determine the LET based on the angle. To determine the LET of the Xenon ion at an angle of 43 degrees, the LET at 0 degree angle is divided by the cosine of the angle. In this case the calculation is as follows:

This is beneficial as it saves time at the cyclotron facility since changing the ion can take 30 minutes or longer. As the old saying goes “Time is money” and the facility has an hourly usage charge. Not only does using the ion angle save some cost but also allows the engineer performing the test to more efficiently use the allotted time. It is important to properly plan ahead and have a good schedule when performing the test to get the best utilization of the time at the facility.

To understand the test process when looking at SETs let’s look at a test procedure. A general procedure for SET testing would be similar to the one used for the AD9246S. For this device the following procedure was used:

- Power up the AD9246S.
- Select the desired ion and desired angle of incidence.
- Turn on the ion beam while observing, monitoring, and logging the power supply currents and recording the transients.
- Turn off the beam when either the specified number of transients is recorded or the fluence reaches 10
^{7}ions/cm^{2}. - Repeat the procedure beginning at step 2 until the AD9246S has been irradiated across the desired range of LETs (
∼
2 MeV-cm
^{2}/mg to 80 MeV-cm^{2}/mg). - Test the remaining AD9246S devices using steps 1 through 5 until the desired number of units have been tested.

The transient responses of the AD9246S during the heavy ion exposure were captured using a Xilinx Spartan 6 FPGA when the SET level exceeded three-sigma of the noise band. The LSBs reflecting board and facility noise were removed from the comparison by applying a digital mask to the output codes of the ADC. For the case of the AD9246S the 6 LSBs were masked. A visual representation showing the error threshold mask (red dotted lines) is given below to illustrate transient errors. The periods outside the mask must be multiple clock cycles in order to be considered a transient. As an example, the first SET is shown as having a length of three clock cycles and a magnitude of greater than 6 LSBs.

This mask was created in the data analysis by performing a logical AND with 0x3FCO (binary value = 11 1111 1100 0000) and each output code. This digital mask represents approximately 1/2 mV of noise with a 2 V_{P-P} input full scale range. For details on each SET test run performed refer to pages 13 – 15 of the AD9246S Single Event Effects Test Report.

This information is used to plot a Weibull curve which shows a visual representation of the SET behavior versus radiation energy level. The more energy levels that are used the more points are available to create the Weibull curve making it more accurately describe the SET response of the ADC. Obviously there is a compromise due to the cost and time associated with utilizing the cyclotron facility to perform the testing. Recall from Table 1 above that multiple LETs are used for testing. This allows the curve below to be generated. This curve shows the LET onset threshold where SETs begin to occur as well as the LET value at which the number of SETs saturate meaning that the number of SETs does not increase with the increase in LET.

The information collected and calculated for the Weibull curve can be used to project the SET performance of the device in various orbits around Earth. The Weibull curve is used to calculate a probability of an SET occurring as well as what the magnitude and length may be for an SET that occurs. Models exist that take the information from the Weibull curve and provide a probability of an SET event dependent upon the orbit that the device will be placed into in a final system application. These models account not only for the orbit but also for any shielding that is utilized in the final application. This just gives one piece of the puzzle.

As we’ve discussed a general first step is SEL testing to look for destructive device latch-up. Our topic here has been SET testing which gives information on the transient upset performance of the device. There are still a few more topics to discuss in order to get the full picture. The next topic we will look at in terms of SEE testing is single event upset (SEU). Stay tuned as we continue to get a clearer picture of SEEs for high speed ADCs. In the meantime, if you’d like to learn a bit more about SETs and Weibull curves here are a few additional resources:

- Failures in Aerospace Applications, Part 1 – The link is for part 1 but this is a multipart series. This is a great blog series to read. It gives some more information on SEEs and in part 6 you can see more on the Weibull curve. (
**Editor’s note: Just do a search on ‘Paolo Scalisi’ on Planet Analog’s site and you will find all the parts of this article**) - Single Event Effect Rate Analysis and Upset Characterization of FPGA Digital Signal Processors – This is an interesting paper from NASA’s archives about SEEs with FPGAs. It includes a bit on Weibull curves as well.
- Engineering Statistics Handbook - Weibull – If you’re really into math and statistics you can have lots of fun reading the details on Weibull at this link.

Copyright © 2019 TechWeb, A UBM Company, All rights reserved.