hundred ns (it depends inversely on the signal charge density in the CH:turret)。 This is comparable with the time spent under the gates during parallel and serial transfer but is not long enough to give the reductions in CTE damage that we have seen。 Hence, we speculate that the clock overlap times (which are much shorter in our case) are also involved。

in Bulk Dark Current

Bulk displacement damage causes the production of dark charge generating states within the depletion  layer。  The effect for low fluences (e。g。 as experienced for MERIS) is a small increase in average dark current and  an  increase in dark current nonuniformity - seen as a 'tail' on histograms of dark current density for each pixel。 It was found that these histograms were similar for all devices。 For example figure 4 shows data for a 100 icm i0opm  epitaxial  thickness CCD and for a 20flcm  25pin  thickness  CCD。   Both  devices  had

similar n-channel geometry and doping。 The received dose was 2 krad of  10 MeV protons (a fluence of 3。6xl0   plc    2)

The plots were obtained for surface inversion conditions and their shape is determined by the bulk damage。

The damage was modelled using the theory described by MafShall et at [10] and their data for nuclear cross sections and damage at 12 MeV (close to our energy of 10 MeV)。 The best fit (Shown in the figure) assumed a damage constant   of

2。4 nA/cjjj2 (2 l°C) per MeV of non-ionizing energy deposited

in the lattice and an active volume of  1500  3  and  a thickness of 4Jim。 These parameters gave a good fit for all 10 MeV dGviOeS and all doses (I, 2, 3 and 4 krad)。 Taking a larger volume (as might be expected for the 100qm device) gave a very poor agreement。 We therefore conclude that, in effect, there is a diffusion length for the migration of vacancy/interstitials and that those created a distance greater than =4Jim away from the depletion layer (or more likely from the n-buried channel where P-V centres will be formed) do not give rise to stable defects that can create dark current。 A device irradiated at several angles of incidence showed increasing nonuniformity (broader histograms) as the angle of incidence increased - but at a rate in line with the increased effective track length within the active volume。

Data for a device irradiated at 100 MeV gave less dark charge nonuniformity (a damage constant of l。3nmcjjj2 at 2l°C when fitted with Marshall et at's 63 MeV data [10] - the

nearest we have available)。 This may be due to the effect of secondary recoils having enough criers' to leave the active volume, as suggested by Marshall et al [10)。 Who also noted a reduced variance for their 63 MeV data。

Knowing the damage constant (K) we can calculate the change in average bulk dark currell)   Jbulk ’

where n is the number of collisions and E the average energy de9osited (ifl MeV)。 The e's and i's refer to elastic and inelastic collisions respectively。 The n values can be calculated knowing the nuclear cross sections, the fluence and

the interaction volume (- 1500qm*)。 For a fluence  of 1。79x10  plc   2   ] krad) and again using the data in [10]:

°*b»lk - 2。4 x (206。 2x1。76xl0-’ + 0。089x0。0765)

- o。 to nmc   2 yr krad at 21°C(10 MeV)。

Note that since we still have a small surface component (even on inversion) we can only infer this value from the nonuniformity distribution rather than measure it absolutely。

Dark Current Density - mean Value, at 21°C (nA/cm°)

Fig 4。 Histograms of dark current density (mean value subtracted) for 100 lcm l00pin and 200。cm 25qm, thick CCD05 devices。 Also shown is a theoretical fit using the method of Marshall et at [10]