Japan Aerospace Exploration Agency JAXA Sitemap

TOP > Report & Column > The Forefront of Space Science > 2012 > When a Computer is Surprised

The Forefront of Space Science

When a Computer is Surprised
| 1 | 2 | 3 |

Ambitiously, I wanted to see the whole waveform. While reviewing my student research, I came up with the idea to realize my wish with just two switches. The result is shown in the left of Fig. 3. This shows for the first time in the world the moment of a shock event on a computation element.

Figure 3
Figure 3. Elucidation of the moment when cosmic rays hit the computation element
Left: Observed result and estimated result of SET noise. Cosmic rays simulated by ultra-short pulse-laser strike the NOT gate computation element fabricated by FD-SOI technology.
Right: Relation between width of SET noise and intensity of cosmic rays. Figures such as 200nm and 90nm indicate nominal dimension of tested elements. Transistor switches are fabricated in the order of nano-meters or one-billionth of a meter. [1] is the result of the bulk-fabricated element, based on experimental data reported in the IEEE Trans. Nucl. Sci., 54(6), p. 2506 (2007) by B. Narasimham et al. It is interesting that the linear relation, which was assumed for the FD-SOI technology, is also found for bulk-type fabrication.

The red dots are the observed result, showing that the element was shocked and its signal dropped at the time of radiation impact. In contrast to Fig. 1, the signal is trapezoid. This is a unique feature of the computation element.

The green line shows the estimated result, based on the method introduced above using a single transistor switch to make the element. The figure shows that both results agree well.

The estimating and measuring methods I developed are very convenient, but there is one insurmountable hurdle. Cosmic rays vary in intensity (i.e., strong/weak). If we successfully measure the SET noise for cosmic rays of a certain intensity, can we also estimate the noise of other cosmic rays with different intensities? The answer is no. Was there any magical calculating formula to forecast the noise based on parameters indicating cosmic rays' intensity? It was hoped that, even if the entire waveform were impossible, perhaps just one parameter would be possible. In fact, I had started research on this matter before the aforementioned technique. In my notes, I started on November 15, 2005. I finally completed the formula in 2009 after filling about 20 A4-size notebooks.

I will omit the detailed formula in this article, but let me explain the most important point. There is a parameter called LET (Linear Energy Transfer) used as an indicator to show the intensity of cosmic rays. I discovered that the width of SET noise was expressed as a logarithmic function of the LET parameter. Thus, in order to double the noise width, cosmic rays of 10 times the power are required. The right of Fig. 3 shows this relation, where the SET noise width is indicated on the vertical axis and the LET is indicated on the horizontal axis. The horizontal axis is plotted logarithmically, so 1, 10, 100, etc., are located with equal space. All the experimental results are plotted as straight lines since they have the nature of a logarithmic function. The formula was established by solving a differential equation based on the physical principle. Thus, I successfully explained the relation between SET noise and cosmic rays both theoretically and explicitly.


In this article, I introduced my research result on SET noise. The research revealed an aspect that even cool computer chips can be surprised. It is somewhat amusing, but frightening for people engaged in spacecraft research. By changing their figures, cosmic rays also fall on the ground, so SET noise may also impact chips on the Earth. It is hoped that this research will also contribute to solving the problem on the ground.

This research was not, of course, achieved solely by myself. I would like to take this opportunity to express my sincere gratitude to the many people who have supported me.


| 1 | 2 | 3 |