The Sensitivity of Thermal Donor Generation in Silicon to Self-interstitial Sinks

Abstract

This work was initially intended to examine the effect of rapid thermal annealing (RTA) on the generation of thermal donors (TD) in silicon in the temperature range 450 to 500C. RTA preanneals made at 1200C in nitrogen were reported 1 to retard TD generation at 450C for about 5 h, and so it was thought interesting to look at the temperature dependence of this retardation. Such an RTA-induced retardation was not found in the present study, however. The difference in the TD kinetics between as-grown and RTA samples was not significant , in accord with other recent data. 2 Instead of the expected retardation we observed another striking effect: it was noticed that the TD concentration was remarkably sensitive to the cooling rate of the sample after the TD-producing anneal steps. This effect was most evident for the RTA-treated samples annealed at 500C, though it was found in as-grown samples too. The effect of annealing conditions on the TD kinetics is the subject of the present work. Experimental We used a set of wafers (0.675 mm thick) cut from the same crystal from adjacent positions and subjected to an RTA treatment at 1250C for 35 s in nitrogen ambient. The crystal, 150 mm in diameter , was grown in the vacancy mode, 3 at the relatively high pull rate of 0.8 mm/min. The oxygen content was 1 10 18 cm 3 (using the calibration factor 3.14 10 17 cm 2), and the carbon content was below the detection limit of 2 10 15 cm 3. The crystal was boron doped to the concentration N B 1.7 10 15 cm 3. Rectangular samples (12 mm long and 3 mm wide) were cut from the central part of the wafers. The thermal donor generation anneals were performed in two different types of furnaces. One of these was an air ambient furnace and the other a vacuum furnace (the residual pressure was 10 3 Torr). The anneals were performed using sequential time steps of 4 h (in most cases). A total duration of up to 80 h was accumulated in this way. In the air furnace each of the sequential anneal steps was followed by one of two different cooling procedures: either quenching (placing a sample on a thick silicon plate) or slow cooling (leaving a sample inside the furnace after switching-off the power). The fast cooling rate was about 60 K/s, the slow cooling rate was about 0.2 K/s. The cooling curves, T(t), were recorded using a thermocouple attached to a sample. After each of the annealing step the samples were slightly lapped so that the sample thickness was gradually reduced, finally down to 0.55 mm. In the vacuum tube furnace the cooling rate was fixed at about 0.8 K/s, intermediate between the slow and fast cooling conditions in the air furnace. The furnace temperature was controlled by a thermocouple, and maintained at prescribed value with the accuracy of 1C for the vacuum furnace and 3C for the air furnace. The TD concentration was deduced from Hall effect measurements at temperatures down to liquid helium. An example of a temperature dependence of the electron concentration n(T) in a TD containing sample is shown in Fig. 1. The electron concentration decreases upon lowering T due to the change in the charge state of the double TD, from doubly charged at room temperature to singly charged at liquid nitrogen temperature. Subsequent decrease in n(T) upon the further lowering of T is caused by the capture of electrons by single-charged donors. The double TD family includes species of slightly different energy levels. The experimental curve is, however, well fitted by the theoretical temperature dependence of n on T (solid line) assuming that the double TD family is represented by one averaged deep level and one averaged shallow level. This fit provides both Thermal donor (TD) generation in silicon at 500C was found to depend significantly on the cooling rate used after sequential annealing steps and on the nature of the ambient (air or vacuum). By performing the anneals initially under some specified cooling rate and ambient, and then changing to a new set of conditions, it was found that the TD concentration relaxed to the value corresponding to the new conditions. These results are well explained by a self-interstitial enhancement of TD generation rate. Self-interstitials are emitted by TD clusters, and their concentration, C i , depends on the efficiency of sinks (sample surface, bulk voids). For vacuum annealing the major sink is the sample surface. For air anneals this sink is "passivated," presumably due to oxidation of the surface and/or by surface contamination, thus leaving only voids to act as self-interstitial sinks. Fast cooling seems to partly passivate voids (presumably by a decoration mechanism), further decreasing the sink efficiency, and therefore increasing C i and the TD generation rate. The quantitative theory of sink-controlled TD generation provides a good description of the complicated experimental kinetic curves. Figure 1. Representative data showing the temperature dependence of the electron concentration deduced from the Hall effect measured down to liquid helium temperature. The sample was annealed at 500C in air (fast cooling mode) for 8 h. The deduced numbers: N 2.45 10 15 cm 3 , deep energy level is 133 meV, shallow energy level is 60 meV, N s 7 10 14 cm 3 .