Measurement Procedure

The measurement procedure for the SEU test differs from the module tests. Fig.  shows the software flow for the single event upset measurements. In the beginning, both hard and soft resets are applied to the APV chips, the registers are loaded over the $\rm I^2C$ bus and pedestals and noise are evaluated as it is done in the module tests. Then, different tests are performed in a loop until an error is detected. These measurements include several thousand software-triggered events with and without internal calibration. The APV voltages and currents are measured at one point in the cycle, and the APV register settings are read back over the $\rm I^2C$ bus. Moreover, the digital optical transceivers (DOT) are tested.

Figure: The principal measurement procedure for the SEU test.

Thus, data are continuously read out with a cycle time of approximately one minute. Each single APV event is subjected to a few checks:

• Is the data frame complete?
• Is the error bit set?
• Is the pipeline address synchronous with the other APVs?
• Do the analog data make sense?

In very rare cases, the entire frame may be missing after a digital SEU in the control logic. When this is not the case, the error bit has to be checked. If it is set, the APV error register can be read over $\rm I^2C$ to distinguish between latency and FIFO errors. The error bit indicates about $81\%$ of all digital single event upsets. Another $15\%$ are detected by matching the pipeline addresses between the APVs on a single hybrid. Since those are run from the same clock and trigger lines, they ought to run synchronously unless a SEU causes an error in the pipeline logic. Approximately $4\%$ of the digital single event upsets flip a bit in the $\rm I^2C$ registers. This condition cannot be safely detected in the output data frame. Either is does virtually no harm (e.g., when the LSB of a register flips), or it affects the signal amplification but not the pedestal output (e.g., by changing of bias voltages or currents in the preamplifier section), or it obviously affects the output (e.g., by shifting the pedestals). A quality check of the analog data should be able to detect at least some of those SEUs. To be sure, one has to read back the $\rm I^2C$ registers and compare them to the set values.

These simple software checks (without $\rm I^2C$ readback) should be implemented whenever the risk of data corruption by SEUs exists. In particular, this appears to be necessary in the CMS data acquisition to discard wrong data.

In addition, the analog data are checked for unexpected hits, which indicate analog SEUs. In case of an error, it is analyzed, counted and logged, and the loop execution is stopped. The program starts over with sending resets and reinitializing the APVs.

The major part of the SEU tests was performed in deconvolution mode with a medium pipeline latency (98) at $-10^{\circ}\,\rm C$. However, data were also obtained in peak mode, at high latency (187) and at room temperature.