Experimental determination of the activation energy (EV D) of vacancy-mediated self-diffusion in silicon crystals is a long-standing issue. Kube et al. [1] studied self-diffusion of the multilayer (ML) structure of isotope Si (20 bilayers of 28Si/29Si) and of a sandwiched (SW) structure (natSi/28Si/natSi) at low temperatures (650–950 °C) with measurements by secondary ion mass spectroscopy (SIMS) and neutron reflectivity (NR). The former and the latter specimens were grown by molecular-beam epitaxy (MBE) and by the chemical vapor deposition method (CVD), respectively. Kube et al. [1] analyzed their data of the SW structure together with those of Bracht et al. [2] at high temperatures (855–1378 °C) and reported EV D as well as E I D (the activation energy of interstitialmediated self-diffusion) to be 3.52 and 4.92 eV, respectively. These values were in good agreement with those (3.6 and 4.95 eV) reported by Shimizu et al. [3] from measurements of diffusion with Raman spectroscopy in the ML structure (20 bilayers of 28Si/30Si). Kube et al. [1] considered these values to be those of an intrinsic crystal since the results of the two groups [1,3] agreed well. To solve the inconsistency that EV D was smaller than that (4.08 eV) of Sb diffusion [4], Kube et al. [1] introduced the temperature dependence of the thermodynamic properties of vacancies, such as the energies of formation and migration, and entropies of formation and migration, which were first proposed by Seeger et al. [5]. Our Comment is on the preferential sites of vacancy formation, i.e., the sites where the formation energy is smaller than that in a perfect crystal. Carbon in the specimens of SW and ML and vacancy clusters in the ML structure are such sites. The carbon concentrations in ML and SW are 3×1018 and 5×1017 cm−3, respectively [1]. Kube et al. [1] assumed that only electrically active impurities have an effect on the charged vacancy formation but offered no explanation on the neutral vacancy concentration which was responsible for the self-diffusion in their specimen. By a quenching experiment, however, we determined the vacancy formation energy in the carbon-doped specimen to be 3.2 eV [6,7] (determined to be 3.08 ± 0.15 eV after reanalysis of data obtained from quenching between 1200 and 1360 °C), much smaller than 3.85 eV [8,9] (determined to be 3.85± 0.15 eV after reanalysis of data obtained from quenching between 1200 and 1360 °C) by our quenching experiment using a high-purity crystal. Nelson et al. [10] calculated the binding energies between various impurity atoms and a vacancy. According to them, the binding energy between the carbon and a vacancy was 0.11 eV, and hence the vacancy formation energy was 3.58 eV since the formation energy in a perfect crystal was estimated to be 3.69 eV by their calculation. Other theoretical estimates have shown that an electrically neutral impurity, such as Sn in Si, has a large binding energy with a vacancy [11]. On the other hand, many vacancy clusters have been detected by positron annihilation studies in MBE-grown structures [12]. The binding energy between the vacancy clusters and a vacancy was estimated to be about 3.2 eV [13]. Hence, the vacancy formation energy from vacancy clusters is 3.2 eV. These results suggest that the data obtained at low temperatures may have been greatly influenced by carbon atoms, not by carbon precipitates, and by vacancy clusters. Hence, the interpretation that the activation energy of vacancy-mediated diffusion was intrinsic would seem to be doubtful. An abnormally small preexponential factor of vacancymediated diffusion seems to be easily explained by the above discussion since the concentration of carbon and/or the density of vacancy clusters are probably involved in the preexponential factor. If we assume that the simultaneous analyses of diffusion data made by Kube et al. [1] and by Shimizu et al. [3] are appropriate, the difference in the preexponential factors between 0.0011 of Kube et al. [1] and 0.0023 of Shimizu et al. [3] can be attributed to the difference in the density of preferential vacancy sites. Incidentally, if the above view that the low-temperature data were influenced by carbon and/or vacancy clusters is correct, there is no reason to analyze the data at high temperatures and low temperatures simultaneously. Plotting numerical data of Kube et al. [1] and those of Bracht et al. [2] (self-interstitial) and Shimizu et al. [3] in larger magnification than Fig. 4 in Ref. [1], we noticed that the relations between log D and 1/T of all data were approximately linear. We attempt to analyze those data, even though it may be beyond the readers’ situation. Figure 1 shows an example. Solid circles correspond to the data of SW. Probably the size of circles should be larger than this plot if we take the experimental error into consideration. The solid line and the broken line correspond to the fitting lines due to the least-squares fit of the data of SW and Eq. (7) in Ref. [1], respectively. In Fig. 2, data of SW and Shimizu et al. [3] are plotted together with the least-squares fitting lines of the solid and broken lines, respectively. Both data agree well. However, as shown by the solid and broken lines, activation energies are slightly different. It is not easy to understand relations among various data from Fig. 4 in Ref. [1]. Hence, in Fig. 3 we show the fitting lines without data points of various data. Solid and open circles correspond to the data at the highest