ABSTRACT
As explained in the previous chapter, different experimental approaches can be considered to estimate the SER of a given device, circuit, or system (Ziegler and Puchner 2004; JEDEC 2006): field tests, accelerated tests, and real-time (or unaccelerated) tests. Field testing consists in collecting errors from a large number of finished products already on the market. The SER value is evaluated a posteriori from the errors experienced by consumers themselves; it generally takes several years after the introduction of the product on the market. Accelerated tests have been largely described in Chapter 6. Accelerated tests use intense particle beams or sources chosen for their capability to mimic the atmospheric (neutron) spectrum or to generate alpha particles within the same energy range as the alphas emitted by radioactive contaminants. This accelerated SER (ASER) method is fast (data can be obtained in a few hours or days instead of months or years as for the other methods), is a priori easy to implement, and only requires a few functional chips to estimate the SER. This allows the manufacturer to perform such radiation tests relatively early in the production cycle. Another major and growing advantage is its capability to quantify from a very large statistic (cumulated number of events) the importance of multiple cell/multiple bit upsets in the radiation response of ICs fabricated in technological nodes typically below 65 nm. But data can be potentially tainted by experimental artifacts (more or less well controlled according to the facility, the experimental setup, or other various experimental conditions). As a direct consequence, ASER results must be extrapolated to use conditions, and several different radiation sources must be used to ensure that the estimation accounts for soft errors induced by both alpha-particle and cosmic-ray-neutron events. We will discuss these issues in Section 7.5.
