As in a previous investigation [6] the bulk  dark current did not depend on the bias applied during irradiation。

Figure 5 shows histograms of dark current density for temperatures in the range 6°C to 40°C。 In each case the histograms can be accurately superimposed by scaling the horizontal axis according to the effective activation  energy。 In this case found to be 0。633 eV。 For example at 10°C we would expect a factor 2。6 reduction in dark current compared with 21°C so the horizontal axis for 10°C data is scaled up by this factor。 Note that the activation  energy  is greater than half the bandgap because the bulk dark current varies as T*

exp(-Eg/2kT) and both the T2 prefactor and the   temperature

dependence of Eg combine to givc a faster change with temperature。 Only the few large dark current spikes due to field-enhanced emission [7, 11] show a different (lower) activation encrgy。

The performance of the MERIS instrument relies on accurate in-flight calibrations of the dark current。 Hence the stability of the dark current between calibrations (nominally oncc pcr day) is important。 A study of proton damaged devices has revealed that some pixels gix'e dark signals which show temporal fluctuations。 Figure 6 shows three types of behaviour: steady (flat) levels which show only‘ system and shot noise (Case C); sharp transitions from one statc to another (sometimes between more than one level) as in cases B or D, and less sharp transitions - or a large number of multiple transitions giving the resemblance of l/f noise (Case

A)。 At room temperature (21°C) a large fraction of pixels can show transitions with amplitudes up to 0。 1 nA/cm 2 and a small number  (around  I in 1,000) have amplitudes up to  0。4

in >2,      Before  irradiation   and  after  cobalt60  gamma

irradiation the pixels show steady values as in C。 Figure 7 shows traces for tbe same pixels shown in figure 6 but at 10°C。 It is seen that the transition amplitudes are decreased and the time constants increased, but the average time in the high and low dark current states are always comparable。

i:irk  st  rem  y- vi u itrm8i zIx›zrG(

Fig。 5 Histograms of dark current density (mean value subtracted) at different temperatures。 The horizontal axes have been scaled so as to normalize to the value the dark current would have at 2l°C assuming an activation energy of 0。63 eV。

The sharp transitions suggest that capture and emission from discrete states is involved。 This has been observed previously [12] however the generation rate from inpidual bulk states is small。 McGrath ct a1 give

to- l5 cm-3 and A = 22。5x22。Spin°, we get ›b '2。6

300K; which  is much  less than the  measured

transition amplitudes。   In addition the capture and  emission

times are of order several minutes, suggesting a small capture cross section =l0‘22 cm° (recall •c — opvyns where •s is the signal density)。   This makes the generation rate  from

equation (3) even less。

The long time constants suggest that a barrier is involved。 For example the defect may lie at the interface between the silicon and the gate oxide。 This is the kind of mechanism proposed for random telegraph signals (RTSs) in small geometry MOSFETS [14, 15J。 Though in that case the field associated with a charged trap ( i。e。 when it has captured an

Fig。 6 Random telegraph signals from a 10 MeV proton damaged CCD at 21°C。 Pixel C shows a steady level with only system and shot noise。 B and D Show sharp transitions between levels and Case A is more like lff noise。文献综述