Analog SEUs

Similar to the flipping of a digital cell, a heavily ionizing fragment can deposit charge on a pipeline capacitor, thus producing a fake hit. The probability of such an analog single event upset is much higher than that of the digital counterpart, since the sensitive area of the pipeline is larger and there is no threshold for the charge deposition.

For practical detection of analog SEUs, a threshold must be defined in order to distinguish true single event upsets from noise. A safe threshold of $40\,\rm\sigma$ (i.e., forty times above the RMS noise level of each channel) has been selected in the SEU tests. This threshold corresponds to charges of approximately $16000\,\rm e$ and $11000\,\rm e$ in deconvolution and peak modes, respectively.

Analog single event upsets of either polarity have been observed, slightly dominated by positive signals. Fig.  shows a typical pulse height spectrum.

Figure: Typical pulse height distribution of analog SEU signals. The central peak shown in red represents the Gaussian noise distribution, which is entirely excluded by the $40\,\rm \sigma$ cut. The calibration for the signal amplitude is approximately $75\,\rm e/ADC$.

The mean amplitude of the positive signals and the measured cross-sections with three different settings are plotted in fig. .

Figure: Mean amplitudes of positive fake hits and total cross-sections measured on the APV25S1.

The analog SEU cross-section with a $40\,\rm\sigma$ cut is a few times higher than its digital counterpart, but is expected to be much higher with a lower threshold. It rises with increasing latency, since the sampled shaper output longer remains in the pipeline and thus each cell is exposed to potential charge deposition for a longer time span.

The measured cross-section in peak mode is higher, because the $40\,\rm\sigma$ threshold corresponds to a lower charge, resulting in a larger number of recognized fake hits. At the same time, the average charge figure decreases due to the lower threshold. Considering these measurement artefacts, both the cross-sections and the amplitudes are compatible between peak and deconvolution modes at similar latency.

Although the probability of an analog single event upset is much higher than the digital SEU cross-section, it does not need any action to be cleared. Since the pipeline is constantly refreshed with a period of the latency, affected cells are automatically overwritten by other samples. Despite the higher cross-section, the number of fake hits is still negligible compared to the overall amount of data.