Continuing Experiments of Atmospheric Neutron Effects on Deep Submicron Integrated Circuits (WP286)

Document ID
Release Date
2.1 English

The Stone of Rosette by Ulrich Schade and Richard Wäsch [Ref 2] defines Rosetta as follows:

Rosetta refers to the crucial breakthrough in the research regarding Egyptian hieroglyphs. It especially represents the “translation” of “silent” symbols into a living language, which is necessary in order to make the whole content of information of these symbols accessible.”

Just as the Rosetta Stone enabled researchers to decode the unsolvable and mysterious Egyptian hieroglyphs by comparing them to the same text written in a known language, the AMD Rosetta experiments link two prior known and well-documented techniques of estimating atmospheric neutron single event upsets (SEUs) with the real effects of atmospheric neutrons on integrated circuits. The known techniques are accelerated testing in a neutron or proton beam and software simulation of the circuit to determine the critical charge a particular node or latch can handle before it changes state. These AMD experiments determine the actual upset rate of AMD FPGAs due to an atmospheric neutron cascade, which resulted from a cosmic ray. With a good understanding of the real effect(s) that these atmospheric neutrons have on today’s integrated circuits, AMD can validate the design and technology choices being used to mitigate these effects.

Predicting atmospheric neutron flux is not an exact science. In the JEDEC89A standard, there is a methodology that uses models and magnetic latitude data to predict the flux at any given location on the earth. The 2005 Rosetta results clearly did not agree with the original JEDEC89 standard, and the committee worked in collaboration with AMD to resolve this issue. Three corrections were made to the JEDEC89A standard:

  • Realization that the proton flux is not insubstantial (it is approximately an additional 7% in San Jose and as much as an additional 32% at Mauna Kea).
  • The attenuation by the building must be more accurately calculated (28% of the flux is lost to the ground floor of a typical Silicon Valley, two-story, tilt-up, concrete structure).
  • A more automated model can be developed to aid investigators (a new prototype web-based tool):

The JEDEC89A revision also proposes a new atmospheric spectral model, based on work done by Goldhagen, et al. [Ref 3].

Thermal neutrons were not considered at the time to have sufficient energy to upset devices. Since the 28 nm technology node, thermal neutrons are now considered, because they do contribute to the overall upset rate.

Alpha particles (contamination) to fabrication and packaging have led to ultra-low alpha manufacturing processes utilized by AMD, and verified by below-ground (shielded) testing.