A. Test Conditions

Irradiations were performed on two devices using the JPL Co⁶⁰ gamma ray source at a dose rate of 10 rad/sec in a standard Pb/Al box. The irradiations were performed at room temperature under dynamic bias, and testing was completed within a few hours, although testing performed several days later showed no change. The devices were tested at dose levels of 1, 3 and 10 krad (10, 30 and 100 Gy).

Room temperature exposure was chosen for convenience, and because the CPS32 is to be operated at near room temperature. As indicated, exposure temperature can affect the hole trapping and interface state generation processes. In particular, at cryogenic temperatures the hole trapping is greatly accentuated, while interface state generation is eliminated. This is one of the distinguishing differences between the APS and the infrared MUX. Dynamic bias was chosen as being representative of operation. Unbiased exposure is expected to be more benign.

Post-irradiation measurements, performed at room temperature, included dark current, flat field illumination, read noise, conversion gain, fixed pattern noise and dynamic range. V_TN and V_TP were measured using the pFET dosimeters and the load transistor of the output amplifier. Dark current, read noise and fixed pattern noise were also measured in an oven at temperatures of -25ºC, 0ºC and +25ºC. Equilibration of the device temperature with the air required 20 minutes, as demonstrated by monitoring of the dark current, and the accuracy of the oven temperature is believed to be better than 1ºC. Alpha particle measurements were made with an Am²⁴¹ source after the 10 krad level.

For the work described here, the clock signals were obtained from a digital output board connected to a computer, operating at 40 kHz. The analog output of the CPS32 was buffered by a low noise amplifier, with a gain of 1x or 10x, and digitized by a 13-bit ADC with a resolution of 1 mV. In the 10x mode the data acquisition system input-referred noise was ~150 µV. All measurements used correlated double sampling.

B. Threshold Shift

The threshold voltage shifts were DV_TN = -1.3 mV/krad and DV_TP = -2.6 mV/krad, consistent with other measurements we have made on this process. The dynamic range, initially ~1.4 V, decreased 2.5 mV/krad. No evidence of subthreshold leakage was seen, and the supply current remained below our measurement threshold of 0.6 mA. These changes are too small to have any practical effect.

The pixel-to-pixel variation of the relaxed reset level, i.e. the fixed offset FPN, was quite large, with a standard deviation s=90 mV. However, this component of FPN is removed by correlated double sampling and doesn’t affect the performance. The shift of the relaxed reset level with radiation was quite uniform across the array, with a standard deviation less than 2.5 mV. This is insignificant compared to the original variability within the frame, and demonstrates that the threshold shifts do not vary from pixel to pixel.

C. Dark Current

As expected, the dark current increased dramatically with radiation. We have normalized dark current to diode the area, rather than to the pixel area, since photocurrent should also scale with diode area. Scaling with pixel area would yield a dark current ~5x lower. As pointed out in Section III, we actually expect the total dose induced dark current to scale with the diode perimeter, rather than the area.

The dark current increased linearly with dose, as shown in Figure 4, and was thermally activated. Figure 5 shows a fit of the entire data set over dose and temperature using the simple equation:

where I₀ is the initial dark current, K is a damage factor, D is the dose, E_a is an activation energy, k is Boltzmann’s constant, and T is the absolute temperature.

The extracted parameters are:
Initial dark current 0.33 nA/cm² (25ºC)
Dark current increase 0.58 nA/cm²/krad (25ºC)
Activation energy 0.50 eV

The activation energy is, as usual, approximately half the bandgap. This is also typical for photodiodes and CCDs. Because of this thermally activated behavior, the dark current performance can be quickly recovered by cooling. The magnitude of the dark current increase is similar to what might be expected for a non-MPP CCD [10]. An additional measurement, made after 8 months of storage at room temperature, showed a small increase in dark current, perhaps as a result of reverse annealing [9].

Figure 6 shows the histogram of the dark current before irradiation, and after 10 krad. The initial distribution is quite broad, in a relative sense, with the ratio of the standard deviation to the mean being s/m=40%. After 10 krad, the absolute distribution is broader, producing a noticeable fixed pattern noise, but the relative variation is only 7%. Considering the 1 krad and 3 krad measurements, it is found that the variance increases linearly with dose.

D. Noise and Conversion Gain

The read noise and conversion gain were measured by the mean-variance method, i.e. as the intercept and slope of the variance against the mean as the illumination level was changed. The variance was taken across ten frames and was pooled over all the pixels, after subtracting the frame averages to remove the effect of variations in integration time and fluctuations in the light level. A conversion gain of 2.4 µV/e- was obtained, consistent with estimates of the diode capacitance, while the read noise was ~100 e-. Neither value changed measurably over radiation or temperature.

E. Random Telegraph Signals

In the course of measuring the noise levels, we observed that the noise level for the dark current, using the pooled variance described above, increased rapidly with integration time, far in excess of the shot noise. A closer examination showed that a few pixels with very high variances were responsible. Time series measurements for these pixels showed the Random Telegraph Signal (RTS) behavior shown in Figure 7. The rapid increase in pooled variance is a result of the fact that the step size is proportional to the integration time. Hence, the variance due to RTS increases with the square of the integration time, in contrast to the variance due to shot noise, which is linear with integration time. Additionally, the number of pixels undergoing RTS transitions increased as the total measurement time increased.

Unfortunately, the Random Telegraph Signals were not identified until after radiation, so it was difficult to determine the effect of radiation on them. However, the overall variances of irradiated and unirradiated parts roughly matched as a function of integration time, and the fraction of pixels showing RTS behavior was ~10% in both cases. This seems to indicate that the RTS did not increase with radiation.

Similar Random Telegraph Signals have been observed in proton irradiated CCDs [11], and are attributed to defects with two configurations, closely spaced in energy. Although we have not yet had the opportunity to study the RTS on the CPS32 in detail, it appears to differ from that reported for CCDs. First, it is present on unirradiated devices, although this could be evidence of residual processing damage. Second, the time scale for the transitions is tens of seconds for the CPS32, as compared with hours for the CCDs. Finally, the transition amplitude appears to be larger for the CPS32, several femtoamps, in contrast to tenths of femtoamps for the CCDs. These differences lead us to suspect that the responsible defects may be interface defects at the gate oxide, rather than the bulk defects presumed to cause the similar effect in CCDs.

Random Telegraph Signals can create a serious noise problem for applications with long integration times, such as imaging, because it makes it impossible to subtract the dark current FPN.

E. Alpha Spectra

Particle detection performance was evaluated by recording the alpha particle spectrum from an Am²⁴¹ source in vacuum, as shown in Figure 8. These spectra represents 2000 frames of 100 ms integration time. Correlated double sampling was used to remove kTC noise and the histogram was computed from the absolute value of the difference of consecutive pairs of frames. This approach removes the offset and dark current FPN component, and is also insensitive to Random Telegraph Signals, except for a single count when a transition occurs within a pair of frames.

Am²⁴¹ alpha particles have an energy of 5.4 MeV and an LET of 0.59 MeV.cm²/mg, and, as shown, produce a signal of 140 mV or ~58,000 e- in the CPS32. We therefore compute a sensitivity of 240 mV/(MeV.cm²/mg). The 1400 mV dynamic range then corresponds to an LET of 6 MeV.cm²/mg.

Several features are evident in Figure 8. First, the low read noise is evident in the fact the nearly two million dark counts are confined to the first 10 channels. Second is the presence of the peripheral hits, which are represented by the uniform background from zero to the peak. As predicted, they number about half the direct hits. Finally we observe that the peak is rather broad, with s/m=7%. It was originally believed that this was due to spread in the source, but a measurement with a surface barrier diode, also shown in Figure 8, demonstrated that this was not the cause. Likewise, the source was small and far enough from the detector to rule out angular effects, although these will occur in space. We now believe the cause is probably pixel-to-pixel sensitivity variation, although this is twice the variation seen in flat field optical illumination.

Since the technique for obtaining good alpha spectra was not developed until after radiation was completed, it is somewhat difficult to judge the effect of radiation. Comparing the alpha spectra of irradiated and unirradiated devices, the differences are small. The alpha peak for the irradiated device is shifted slightly (3%) to the left, but this is probably the result of gain variations. The peak for the irradiated device is also somewhat noisier and wider. Most interesting, however, is that the peripheral hits are down to only 35% of the direct hits. This may again be part-to-part variation, or it may represent an increase in the surface recombination velocity under the reset gate due to interface states.