Edward Petersen

Rate prediction for single event effects-a critique

The authors review various single event effects (SEE) testing and rate prediction methodologies and recommend standard approaches. This discussion is limited to single event upset (SEU) rate prediction for direct-ionization-induced effects. The standard approach being recommended is based partially on a different way of viewing the results of SEU cross-section measurements. The measurements are not measuring a distribution of cross-sections. They are measuring a distribution of device sensitivities, due to differences of sensitive region critical charges and to differences of charge collection. The linear energy transfer (LET), at which 50% of the cell population upsets, corresponds to the charge deposition necessary to upset the median cell in the circuit array. The threshold LET corresponds to the most sensitive region being hit in its most sensitive location, and does not represent the entire array. The shape of the cross-section curve is described by an integral Weibull distribution. The upset rate for a device should then be calculated using the differential rate of each sensitive region, combined with an integral weighting given by the Weibull distribution that describes the measured cross-section curve. >

Calculation of Cosmic-Ray Induced Soft Upsets and Scaling in VLSI Devices

Progression of VLSI circuitry to smaller feature sizes raises questions about an increased severity of the cosmic ray upset problem. In this paper we present a simple method of calculating cosmic ray upset rates. We compare the results of this method to results of an exact calculation and apply both methods to the prediction of upset rates as device feature sizes are scaled to submicron dimensions. The exact calculations are presented for several environmental predictions. We then discuss upset critical charge as a function of feature size. We consider upset rates versus scale parameter as a function of device size and critical charge. We conclude that upset rates do not increase catastrophically as devices scale down, but that the problem will be serious for all technologies. We also conclude that devices with small feature sizes will be susceptible to upsets by proton induced reactions, so that they will have serious problems in the proton radiation belt.

Suggested Single Event Upset Figure of Merit

This paper examines a number of concepts that are connected, directly or indirectly, with the problem of assigning a single event upset figure of merit to a specific oevice. Single event rates depend both on device and circuitry, through the critical charge requisite for upset, and upon the device geometry and technology, which determine the target size and charge collection capability. Each of these factors must be taken into account when determining device susceptibility. Upset rates in space additionally depend on the environment. Device response in trapped proton belts and in the cosmic ray environment is sufficiently oifferent that a single susceptibility measure is inadequate. We conclude that devices should be characterizea by a proton susceptibility, and by an upset rate in a reference cosmic ray environment. We present a simple expression, based on laboratory measurement, that approximates the cosmic ray upset rate ano propose it as a figure of merit. Calculated and measured values for upset rates are sensitive to several factors. The field funneling effect is known to increase both the magnitude of collected charge and the effective sensitive circuit volume during single events relative to the static parameters. Thus, experimental sensitive area measurements obtained from single event data exceed values predicted from inspection of static circuit layout configurations. Also, more charge is collected from any specific event than is predicted by using depletion region extent to determine charge collection volumes. This effect must be included when critical upset charge values are determined from experimental upset thresholds.

Proton Upsets in Orbit

This paper presents a method of predicting proton-induced single event upset rates in spacecraft RAMs. The approach uses a sensitivity parameter A, determined from one or more experimental measurements of upset cross sections made at any proton energy above threshold. Parameter A uniquely determines a curve for the energy dependence of the upset cross section. This curve can be combined with the proton spectrum at the RAM to predict its upset rate. Predicted upset rates for 600 circular orbits are presented.

Geometrical factors in SEE rate calculations

Examines a number of possible geometrical effects that may show up in either upset measurements or upset calculations. The geometrical effets are with respect to a number of unusual experimental measurements, and an attempt is made to fit these results into a common set of concepts. In most cases, the results will not be decisive and there will still be room for alternative analysis. In some of these cases, it may be necessary to perform detailed charge collection or microbeam experiments in order to reach closure. However, the authors believe that the concepts and questions that they introduce are fundamental for a complete understanding of single event upsets in modern devices. In particular, they continue to maintain that the basic upset cross section curve can be represented by a single smooth curve. They summarize the knowledge of the funnel effect and indicate approaches for including the funnel in upset rate predictions. >

Two parameter Bendel model calculations for predicting proton induced upset (ICs)

The two-parameter model of W.L. Bendel and E.L. Petersen (1984) represents an improvement over the existing single-parameter model in terms of the goodness of fit to actual proton upset data. It especially gives a better fit to the data from devices with small feature dimensions, which ultimately leads to a more accurate proton error rate prediction. Small feature sized devices have a proton upset energy dependence that cannot be accurately described with the one-parameter model and only one data point. There are no substantial differences in proton error rate predictions from one- and two-parameter approaches for older devices with larger feature sizes. >

Soft Errors Due to Protons in the Radiation Belt

This paper examines the problem of soft errors in semiconductor devices caused by the protons in the radiation belts. The errors can be produced by a variety of nuclear reactions in silicon. A previous paper presented a calculation of the likelihood of some of these reactions. This information can be combined with knowledge of the proton environment in order to estimate the upset rate for various devices in spacecraft. This paper discusses the proton environment, the effect of spacecraft shielding, the various proton induced reactions in silicon, the calculation of soft error sensitivity, and the soft error rate in a representative satellite.

Predictions and observations of SEU rates in space

This paper summarizes the available data on the observation and prediction of SEU rates in space. It considers three questions: 1. How good can a prediction be? 2. How bad can a prediction be? 3. How does the quality of the prediction depend on the type of orbit? The paper considers one hundred and twenty-six reports of predicted and observed rates. These include updated predictions for the CRRES devices. The analysis then excludes solar particle events, single event burnout, cases with poor statistics, and cases that are essentially duplicates; leaving 77 comparisons. The heavy ion predictions based on the CREME environments and the proton predictions based on the AP8 environments are both very successful for their basic environments, but are less accurate for low earth orbits (LEO). The quality of the results depends strongly on whether the predictions are based on tests with flight parts or with generic parts. The quality also depends on the use of the proper shielding around the part. The results appear consistent with suggested modifications in these environments based on recent space measurements. The methods that are used for upset rate predictions appear to be adequate for the current generation of devices.

Approaches to proton single-event rate calculations

This article discusses the fundamentals of proton-induced single-event upsets and of the various methods that have been developed to calculate upset rates. Two types of approaches are used based on nuclear-reaction analysis. Several aspects can be analyzed using analytic methods, but a complete description is not available. The paper presents an analytic description for the component due to elastic-scattering recoils. There have been a number of studies made using Monte Carlo methods. These can completely describe the reaction processes, including the effect of nuclear reactions occurring outside the device-sensitive volume. They have not included the elastic-scattering processes. The article describes the semiempirical approaches that are most widely used. The quality of previous upset predictions relative to space observations is discussed and leads to comments about the desired quality of future predictions. Brief sections treat the possible testing limitation due to total ionizing dose effects, the relationship of proton and heavy-ion upsets, upsets due to direct proton ionization, and relative proton and cosmic-ray upset rates.

Nuclear Reactions in Semiconductors

Soft upsets in semiconductor memory devices can be produced by charged particles produced in nuclear reactions in the semiconductor or in its surrounding materials. We have calculated the particle production cross sections for incident neutrons and protons in various semiconductor materials in the energy range of 5 to 75 MeV. The common semiconductor elements and compounds all have approximately the same alpha particle production. There is also appreciable proton and heavy ion production which under some conditions may cause upsets.