Laser Simulation of Single-Particle Effects

Transcription

1 IECE?.ANSAC I lons ON NUCLEAR SCIENCE, VOL. 39, NO. 6, DECEMBER 1992 Laser Simulation of Single-Particle Effects 1647 INTRODUCTION C.A. Gossett, B.W. Hughlock, and A.H. Johnston High Technology Center Boeing Defense and Space Group Seattle, WA Earlier work has indicated that pulsed picosecond lasers can produce charge tracks in semiconductors which are similar to the charge tracks produced by heavy ions[ 1.21, and that in many cases this allows a laser to adequately simulate single-particle effects with greater convenience and lower cost compared to particle accelerators. Although there are similarities in the charge generation from lasers and heavy ions, differences in the radial characteristics of the resulting ionization tracks can affect the equivalence between them. Several factors make it difficult to calibrate laser intensity for testing or screening purposes including optical reflection losses and the spreading of the light within the semiconductor due U, refraction. In addition, the carrier density in the ionization track produced by the laser can be two or more orders of magnitude less than that for an ion back This may have serious implications on the suitability of the laser to reproduce the charge funnelling process characteristic of heavy ions. Diffused charge may also be greater when a laser is used due to increased penetration depth compared to accelerator produced ion beams. Although it is possible to calibrate a laser with heavy-ion experiments on a specific set of devices, the issue of track density and its effect on prompt charge has not been thoroughly examined This paper examines laser simulation in more detail, investigating the effects of doping density and charge collection depth on funnelling and charge collection, and accounting for the spreading of the focussed laser beam within the silicon material. Calculations using the device simulation code PISCES are compared for heavy ions and lasers to investigate the accwacy of laser simulation of single-particle effects. Experimental charge collection measurements with heavy ions are used to establish the accuracy of the PISCES simulations. LASER PROPERTIES Although lasers with various wavelengths have been used for single-particle upset simulation, a pulsed picosecond Nd:YAG laser with wavelength 1.6 pm is probably best suited for realistic simulation results. At a wavelength of 1.6 pm the absorption coefficient in silicon is approximately 4cm-l at room temperature and hence the l/e penetration depth is roughly 25 pm. By contrast, lasers with shorter wavelengths, for example dye lasers (.9 pm) or frequency doubled Nd:YAG lasers (.53 pm), have quite shallow penetration depths. Figure 1 compares d e r generation of these lasers as a function of depth into silicon material. The penetration of a.53 pm laser is extremely shallow and most of the energy is absorbed very near the surface. The.9 pm dye laser has a penetration depth of approximately 19 pm which, as will be shown later, is insufficient for reproducing single-particle effects in lightly doped materials where charge may be collected over depths of 3-4 pm. The usual method of applying lasers for simulation of single-particle effects in integrated circuits involves using a microscope objective to focus the beam to a small spot. This provides a way to determine the location of the laset pulse on the circuit, However, due to the relationship between numerical aperture and resolution, a small spot size can only be obtained by using objective lenses with large incident angles. Specifically, the numerical aperture, N.A., equal to the product of the index of refraction and the sine of the angle of incidence, determines the resolution, R through the relationship: For example, the minimum spot size would be approximately 1.6 pm in diameter for a numerical aperture of.4. Thus when focussed to a small spot, the laser beam enters the device a relatively large angles. This has two important effects. First, it increases surface reflection losses and makes it much more difficult to calculate reflection losses since the simple formulas for normal incidence no longer apply. Second, it causes the laser beam to diverge with increasing distance into the material due to refraction. E T.6 - normalird t.ud.c o- I I lo Dirt.no (lull1 Figure 1. Laser absorption as a function of depth into silicon for wavelengths 53..9, and IEEE

2 1648 For normal incidence, surface reflection will be about 3% at an aidsilicon interface. In practice, there is a surface layer of Si2 which further complicates surface reflection. The thickness of the Si2 layer may cause constructive or destructive interference, which also depends on the angle of incidence. The net effect is a relatively large uncertainty in surface reflection. This must be measured or realistically calculated in order to get reasonable agreement between heavy ion and laser results. Note that surface reflection is also affected by beam polarization which is an additional complication. The high index of refraction in silicon causes the angle at which the laser diverges within the material to be much lower than the incident angle. Nevertheless, for a highly focussed spot at the surface, the spot sue increases substantially with incmsing distance into the material. For a numerical aperture of.4 as above and spot size of 1.6 pm, the beam diameter will have increased to 2.7 pm diameter at a depth of 5 pm. Laser charge generation is affected also by the energy and energy density. Calibration methods must take into account all of these factors: refraction, reflection, and energy density. HEAVY ION EXPERIMENTS Charge collection measurements were performed at the University of Washington tandem Van de Graaff accelerator using the ions listed in Table 1. The test devices were connected through a charge sensitive preamplifier to a pulse shaping amplifier and then to a multichannel analyzer. The charge collection measurements were calibrated with measurements of the energy deposited by the incident particles in a silicon surface barrier detector located adjacent to the test devices. The test devices were originally designed as large area latchup test structures. For the measurements reported here they were connected as large reverse biased diodes. Charge collection measurements as a function of bias were obtained for junctions diffused into the substrate and for junctions diffused into a well. The contacvsubstrate junctions were rectangular rings 4 pn wide with outside dimensions 18 pm by 96 pm. The contacvwel1 junctions had rectangular dimensions 16 pm by 28 pm. The devices were fabricated with 2 micron p- and n-well processes and arrays of 1449 devices were connected in parallel on each die to provide large sensitive areas for charge collection measurements. Realistic doping profiles were obtained from spreading resistance measurements. Table 1. Ion Parameters Ion Energy Range Incident LET m) mevcm2/m g) B C F Si Cl The charge collected in an n+/p-substrate junction as a function of bias for the ions listed in Table 1 is shown on the left side of Fig. 2, while that measured for the p+/n-well junction is shown on the right. Note the factor of two difference in vertical scales. The well doping of the device was 5~1~7 cm-3 while the substrate doping was 7x1Ol4 cm Q P o-j Reverse Bias (Volts) Reverse Bias (Volts) Figure 2. Charge collected for ions listed in Table 1 as a function of applied bias. Left Charge collected in a n+/psubstrate junctions. Right: Charge collected in a p+/n-well junction.

3 The voltage dependence of the collected charge for the n+/p-substrate has a similar shape as that observed by M c h and Oldham [3,41. The charge collected in the p+/n-well junction is much less voltage dependent. In addition, the charge collected in the p+/n-well junction scales linearly with ion linear energy transfer (LET), while the charge collected in the n+/p-substrate junction has a much less straightforward LET dependence. This is illustrated further in Figure 3. The incident energies of the carbon and boron ions listed in Table 1 were chosen to compare results for ions with the same LET but with ranges in silicon differing by nearly a factor of 2. The amount of charge collected in the n+/p-substrate junction is clearly sensitive to the effect of ion range while that collected in the pe/n-well junction is nearly independent of range mobility were applied. Of particular importance is the inclusion of carriercarrier scattering in the version of PISCES used. Carrier-carrier scattering has large effects on the radial and the dependent diffusion of charge, particularly for the very high carrier concentrations found along the ionization tracks produced by the charged particles or laser beam. Indeed the lack of success of the earlier work of Knudson and Campbell [6] to reproduce measured charge collection waveforms may have been due to the absence of caniercarrier scattering in the version of the code used by those authors. Path of incident particle - t I 1 I I I ' I 1.5 : n LET (MeV-cm2/mg) Figure 3. Charge collected in pt/n-well and nt/psubstrate junctions for ions of varying LET. See Table 1. Applied bias of 5 volts. MODELING The semiconductor device simulation code PISCES was used to examine the suitability of laser simulation of singleparticle effects by comparing the charge collection from a moderately heavy particle with that from a laser. The continuity equations for holes and electrons and Poisson's equation are solved self-consistently in PISCES for the electrostatic potential and the electron and hole concentrations. The geometry chosen for the modeling reported in this work is shown in Figure 4. The geometry was chosen to be cylindrically symmetric about the ionization track. For this case, the 2dimensional calculation was effectively 3-dimensional. The version of PISCES used was that from Technology Modeling Associates version 933 [5]. Auger and Shockley- Read-Hall recombination, as well as band gap narrowing were included. Models for both field and concentration dependent Doping densities: n+ contact = IO1' cma3 p- substrate = 7 x IOl4 cm-3 (Case 1) 7 x 115 cm-3 (Case 2) 7 x 116 ~117.~ (Case 31 Figure 4. Device geometry used for charge collection modeling. The results of measured charge collection in the n-well process test structure are compared with PISCES predictions in Figure 5. The finite ion range and linear energy transfer dependence on depth were included in the calculations. Quite good agreement is observed, with PISCES reproducing the light ion results quite well and overestimating the charge collection for the heavier ions by approximately 2%. Note that the simple cylindrical geometry described above was used for the calculations although the actual geometry was a complicated 3-dimensional one. Having established the success of PISCES in reproducing measured charge collection results, one may now apply PISCES simulations to compare charge collection from heavy ions and lasers. Shown in Figure 6a are contours of electron density and electrostatic potential calculated for ion and laser tracks 1 picoseconds after incidence. For both the ion and law tracks the LET as a function of depth into the silicon was assumed to be constant. The incident laser beam spot size was assumed to be 1.5 gm in diameter with a divergence of 7' in the silicon. The device modeled was a -substrate device with a rather low doping density of 7x1p4 ~ m - ~ As. mentioned earlier, cylindrical symmetry about the ion or laser track was assumed. These tracks appear along the left edges of the

4 165 figure. The dashed curves extending primarily in the vertical direction are contours of electron densi The central track density for the ion track is roughly 2 ~1~9 while for the laser it is more than two orders of magnitude lower, approximately I x ~ O ~ ~ The effect of the divergence of the laser beam within the silicon is evident. Charge (PC I t PISCES 3*5 Measured 1. 1 that funnelling was exhibited by lasers with extremely high LET, greater than 5 MeV-cm2/mg. The present results are in agreement with that work. That is, for extremely high LET where the central Carrier density in the ionization track is much larger than the background doping density, charge funnelling with a laser is expected in spite of the very large track diameter limited by focussing of the laser beam. However, looking ovec a broader range of LET and doping density, particularly for the high doping densities required for aggressively scaled devices, this will clearly not be the case. For high doping density or low LET the laser does not reproduce heavy ion results primarily due to the effect of substantially reduced d e r density in the ionization track. I I I I Incident Energy IMeV) Ions: B C F Si CI Figure 5. Comparison of calculated and measured total charge collection for boron. carbon, fluorine, silicon and chlorine ions. See Table 1. The solid cuives in Figure 6 =present contours of constant electrostatic potential. One observes clear evidence of funnelling along the ion and laser tracks, that is, a strong distortion of the potential contours along the track. Contours of electron density and electrostatic potential.5 nanoseconds after incidence are shown in Figure 6b. At these times the funnelling has essentially collapsed and the induced charge density along the particle or laser track has diffused radially. The results of PISCES calculations indicate that the prompt charge collected for the laser is in good agreement with that from the heavy ion while the diffused charge collected from the laser track is reduced by approximately 2%. Thus for the low doping density device modeled here, the laser does appear to repmduce the funnelling effect which would be observed for a heavy ion, and the prompt charge collected is in good agreement. In contrast, funnelling is not observed for the laser simulation for high doping density. For example, electron density and potential contours 1 picoseconds after a laser strike are shown in Figure 7 for a device with substrate doping of 7x1Ol6 In this case no funnelling is observed for the laser track. The results of the present work indicate that in certain cases, for low doping density or high effective LET, that a laser of an appropriate wavelength can be used to simulate single-particle effects in silicon. In previous work, Buchner, et al[7], concluded from total charge collection measurements Dntsnca Imiuonsl Figure 7. Calculated contours of electron density and tential for laser track 1 ps after incidence. Substrate doping 7xlOG :3 Another important factor in comparing heavy ions and laser simulation is ion range. Most experimental work on charge collection has included use of ions with ranges of pm [3,4,8], with the assumption that the prompt charge collected would be unaffected as long as the ion range exceeded the funnelling depth. However the range of ions encountered in space environments and the effective range of a 1.6 pm wavelength laser in silicon far exceed the funnelling depth. Figure 8 shows PISCES calculations of prompt and total collected charge for ions with various ranges, but with constant LET of 3 MeV-cm2/m along the particle range. A subslrate doping of 7x1Ol4 cm $ for a diode reverse biased at 5 volts was assumed. The modeling mults indicate that the prompt charge does not saturate until the range of the ion exceeds by about 5% the funnelling depth estimated from the ratio of carrier mobilities. In order to reproduce the total charge that would be collected, an ion range roughly twice the estimated funnelling depth would be required for the doping density and LET assumed here. These effects become more

5 Particle Laser V I i :. I Dirtancm tmlcrons) Distance lmicronrl Particle Distance Inriuons) 7 t o" / , ,.-...( A Distance knkm"i Figure 6. Calculated contours of electron density in electrons/cm3 (dashed curves) and potential in volts (solid curves) for ionization tracks (along left edges, axes of cylindrical symmetry) from LET = 3 MeV-cm2/mg "particles" and laser beams. Substrate doping 7x114 cm3. (a) Contours 1 ps after incidence. (b) Contours.5 ns after incidence.

6 1652 pronounced for higher LET. Thus an important conclusion of the present work is that care must be taken in selecting laboratory ions with sufficient range for simulation of singleprtrticle effects in a space environment where the ion ranges can be quite long, or for calibration of laser intensity in a specific device. Charge (PC) 1. I I.8 L O Total Prompt t Funneling depth based on mobility ratio I I I I I J ' Range (pm) Figure 8. Calculated pmmpt and total charge as a function of range for "particles" with constant LET = 3 MeVcmqmg. The modeling and heavy-ion charge collection data discussed above have serious implications for attempts to calibrate the intensity or effective LET of a laser relative to heavy ions. The results illustrated in Figures 2 and 3 indicate that charge collection for the n+/psubstrate junction is clearly sensitive to factors such as ion range and applied bias, while charge collection in the contact/well junction appears to be limited by the well depth and is linear with incident LET. Therefore one recommendation of the present work is that charge collection for structures embedded in a well should be used to calibrate laser intensity compared to heavy ions. Complication of laser versus heavy-ion calibration due to the inherent differences in the carrier density in the ionization track, divergence of the laser beam within the material, differences in the depth dependence of the charge deposition and the effects of ion and effective laser range on totai and prompt charge collection may be reduced by calibration using structures embedded in wells. Due to the limitation of charge collection depth by the well itself, the effects of the differences mentioned above are significantly reduced. Note that the "ion shunt effect" [9,1] has neither been observed experimentally nor modeled in this work. Based on the results discussed above, for ions and devices where the ion shunt effect is not observed, one would expect that the above conclusions and recommendations regarding calibration of laser effective LET using junctions embedded in wells would remain valid. Further modeling and experimental work may be needed to quantify the importance of the shunt effect in I calibration of effective laser LET, particularly for devices in which the shunt effect is observed with ions. CONCLUSIONS The semiconductor device simulation code PISCES was used to examine the suitability of laser simulation of singleparticle effects in simple silicon diodes. For low doping density devices or high effective linear energy transfer, laser simulation of single-particle effects appears to be valid. However, care must be taken to choose ions with sufficiently long range in order to calibrate the laser intensity and corresponding effective LET. Calibration of effective laser LET compared to heavy ions should be performed by charge collection measurements with structures embedded in wells where differences between heavy ions and lasers due to the density of the ionization track, variation of charge deposition with depth, and the dependence of collected charge on effective range will be minimized. For high doping density or low effective LET, lasers do not reproduce the charge funnelling effects observed with heavy ions. The results of modeling predictions mdicate that ions of range much larger than the expected funnelling depth for a device should be used for laboratory simulation of space environment particles which may have quite long ranges in silicon. REFERENCES [l] S.P. Buchner. D. Wilson, K. Kang, D. Gill, and J.A. Mazer, "Laser Simulation of Single Event Upsets", IEEE Trans. Nucl. Sci., Vol. 34, pp , December [2) A.K. Richter and I. Arimura, "Simulation of Heavy Charged Particle Tracks Using Focused Laser Beams", IEEE Trans. Nucl. Sci., Vol. 34, pp , December [3] F.B. McLean and T.R. Oldham, "Charge Funnelling in N- and P- Type Si Substrates", IEEE Trans. Nucl. Sci., Vol. NS-29, pp , December [4] T.R. Oldham. F.B. McLean, and J.M. Hartman. "Revised Funnelling Calculations for Heavy Particles with High W&", IEEE Trans. Nucl. Sci.. Vol. NS-33, pp , December [5) Technology Modeling Associates PISCES version 933, Technology Modeling Associates, Palo Alto, CA. [6] AB. Knudson and A.B. Campbell, "Comparison of Experimental Charge Collection Waveforms with PISCES Calculations", IEEE Trans. Nucl. Sci., Vol. 38, pp December (71 S. Buchner, A. Knudson, K. Kang, and A.B. Campbell. "Charge Collection from Focussed Picosecond Laser Pulses", IEEE Trans. Nucl. Sci.. Vol. 35, pp December [8] R.S. Wagner, N. Bourdes, J.M. Bradley, C.J. Maggiore, AX. Knudson, and A.B. Campbell, "Alpha-, Boron-, Silicon-, and Iron- Ion-Induced Current Transients in Low Capacitance Silicon and GaAs Diodes", EEE Trans. Nucl. Sci.. Vol. 35, pp , December

7 A.R. Knudsen, A.B. CampbeU, P. Shapiro, WJ. Stapor. E.A. Wolicki, EL. Petersen. SE. Diehl-Nagle, J. Hauser, and P.V. Dressendorfer, "Charge Collection in Multilayer Structures", IEEE Trans. Nucl. Sci.. Vol. NS-31, pp December [lo] A.R. Knudsen. A. B. Campbell, J.R. Hauser, M. Jesse, WJ. Stapor,and P. Shapiro, "Charge Transport By the Ion Shunt Effect", IEEE Trans. Nucl. Sci., Vol. NS-33. pp. 156, December 1986.