Help & Methodology Reference
The REDW Radiation Effects Distributions tool is a web-based analysis platform that visualizes compiled radiation test data from more than three decades of the IEEE NSREC Radiation Effects Data Workshop (REDW). The dataset spans 1992–2025 and encompasses approximately 7,900 individual test records, providing a broad statistical picture of Total Ionizing Dose (TID) hardness and Single-Event Latchup (SEL) immunity across a wide range of commercial and radiation-hardened device types.
The tool is organized into four tabs: TID Hardness (survival distribution curve of TID pass levels), SEL Immunity (distribution of SEL LET thresholds), Statistics Table (percentile statistics broken down by device type), and Parts List (filtered table of individual test records). Interactive filtering controls allow the user to slice the data by part grade, publication year range, minimum TID threshold, and device type.
The primary purpose of the tool is to help radiation effects engineers develop intuition about the statistical landscape of radiation hardness across device categories — supporting early design trades, parts screening, and risk assessment before device-specific test data is available.
The Radiation Effects Data Workshop (REDW) is an annual compendium of radiation test results published as part of the IEEE Nuclear and Space Radiation Effects Conference (NSREC). Each year, the radiation effects community contributes test results on commercial, military, and space-grade electronic devices to the workshop, which are compiled and published in the conference proceedings.
REDW papers typically report TID test results (pass/fail levels in krad(Si)), SEE test results (LET thresholds for latchup, upset, and other single-event phenomena), and sometimes displacement damage data. The workshop has been published annually since 1992, making it the most comprehensive longitudinal dataset of radiation test results available to the community.
The data compiled in this tool is drawn from the TID and SEL sections of the annual REDW publications. Each record corresponds to a specific device tested at a specific facility, with results as reported by the original authors. The dataset reflects the collective testing efforts of government laboratories (NASA, DoD, DOE), aerospace companies, and university research groups over more than 30 years [1].
The TID Hardness tab displays a survival distribution curve showing the fraction of tested devices that remain functional at or above each TID level. The horizontal axis is TID level in krad(Si) on a logarithmic scale, and the vertical axis is the fraction of devices surviving (from 0 to 1).
At any point on the horizontal axis, the height of the curve indicates what fraction of devices in the selected population were tested to at least that TID level and found to still be functional. For example, if the curve reads 0.60 at 50 krad, it means 60% of the tested devices in the filtered set passed TID testing at 50 krad or higher.
The curve is monotonically decreasing: as the TID level increases, fewer devices survive. The curve starts at 1.0 (all devices survive at 0 krad) and decreases toward zero at very high TID levels. The shape of the curve reveals the distribution of hardness across the device population — a steep drop indicates a narrow range of failure thresholds, while a gradual decline indicates wide variability in radiation hardness.
A critical subtlety of the TID data is that most test results are right-censored. When a REDW paper reports that a device "passed 100 krad," this means the device was functional at 100 krad — but the actual failure threshold may be significantly higher. The test was stopped (censored) at 100 krad, so the true failure point is unknown but is at least 100 krad.
This right-censoring means that the survival curve displayed by the tool is a lower bound on the true survival distribution. The actual fraction of devices surviving at any given TID level is at least as high as what the curve shows, and likely higher. This is especially important at the upper end of the TID range, where many devices may have been tested to facility or program limits rather than to actual failure.
Several factors contribute to right-censoring in REDW TID data:
Test facility limits. Many test facilities have practical dose limits (e.g., 100 krad or 300 krad) beyond which testing is not performed. Devices that would survive to higher levels are recorded at the facility maximum.
Program requirements. Mission-driven testing often stops once a device exceeds the mission TID requirement with adequate margin. A device tested to "pass 50 krad" for a 25-krad mission may actually survive to 200 krad or beyond.
Dose-rate considerations. TID testing per MIL-STD-883 TM1019 [10] is typically performed at high dose rates (50–300 rad(Si)/s). Some devices, particularly bipolar technologies, exhibit Enhanced Low Dose Rate Sensitivity (ELDRS), where degradation is more severe at the low dose rates characteristic of the space environment [8, 9]. Conversely, some MOS devices show time-dependent recovery effects that make high-dose-rate testing conservative.
The SEL Immunity tab displays the distribution of Single-Event Latchup (SEL) LET thresholds across the tested device population. The horizontal axis is LET threshold in MeV·cm²/mg, and the vertical axis shows the fraction of devices confirmed to be SEL-immune at or above each LET level.
The curve shows, for each LET value, the fraction of tested devices that exhibited no latchup at that LET or higher. A value of 0.70 at 40 MeV·cm²/mg means that 70% of the tested devices in the filtered population showed no SEL at LET values of 40 MeV·cm²/mg or above.
Like the TID curve, this is a monotonically decreasing survival-type function. At low LET values, nearly all devices are immune (the curve is near 1.0); as LET increases, the fraction of immune devices decreases as more parts begin to exhibit latchup susceptibility.
A prominent feature of the SEL distribution is a concentration of data points near 120 MeV·cm²/mg. This is not a physical phenomenon — it is an artifact of test facility capabilities. The maximum LET achievable at most heavy-ion test facilities (using ions such as gold or uranium at typical energies) is approximately 80–120 MeV·cm²/mg. When a device shows no latchup at the maximum available LET, it is recorded as "SEL-immune to ≥120 MeV·cm²/mg" (or similar).
This means that a reported SEL LET threshold near 120 MeV·cm²/mg should be interpreted as "no latchup observed up to facility maximum" rather than "latchup onset occurs near 120 MeV·cm²/mg." These data points are effectively right-censored: the true SEL immunity threshold is at least as high as the tested maximum, and potentially much higher (or the device may be truly SEL-immune at all LET values) [6, 7].
For mission applications, the SEL LET threshold is compared against the expected particle environment. In geosynchronous orbit behind typical shielding, the effective LET cutoff for galactic cosmic rays is approximately 30–40 MeV·cm²/mg. Devices with SEL thresholds above this level are generally considered SEL-immune for that environment. However, worst-case environments (solar heavy-ion events, minimal shielding) can produce effective LET values up to approximately 100 MeV·cm²/mg.
The Statistics Table tab provides percentile breakdowns of TID and SEL data for each device category. Understanding the statistical methods used to compute these values is essential for proper interpretation.
The tool reports P10, median (P50), and P90 values for both TID pass levels and SEL LET thresholds within each device category. These percentiles are computed from the empirical cumulative distribution of the reported test values.
P10 is the 10th percentile: 10% of tested devices had results at or below this value, and 90% had results at or above it. This represents a conservative, worst-case estimate of what to expect from a randomly selected device in that category.
Median (P50) is the 50th percentile: half the tested devices had results above this value and half below. This represents the "typical" device in the category.
P90 is the 90th percentile: 90% of tested devices had results at or below this value. This represents the high end of the distribution — most devices in the category will not exceed this level.
Because TID data is right-censored (reported pass levels are lower bounds on true failure thresholds), standard percentile calculations applied directly to the reported values will underestimate the true distribution. The survival curves displayed in the tool use a Kaplan-Meier-style approach adapted for this right-censored structure [2, 5].
In classical Kaplan-Meier analysis, each observed event (failure) reduces the survival estimate, while censored observations (devices that did not fail before testing stopped) contribute to the risk set but do not directly reduce the survival estimate. The result is a step-function survival curve that properly accounts for the incomplete information in censored observations.
The reliability of percentile estimates depends strongly on sample size. For device categories with large numbers of records (e.g., Op-Amps with hundreds of records), the distribution is well characterized and percentiles are stable. For categories with fewer records (e.g., Sensors or MCU/Processors), percentiles have wider confidence intervals and should be treated as rough estimates only.
The Statistics Table displays the number of records (N) for each category and test type. As a rough guideline, percentile estimates from categories with N < 20 should be treated with caution, and categories with N < 10 should be considered anecdotal rather than statistical.
The grade filter allows the user to restrict the dataset to specific part classifications. Understanding these categories is important for meaningful comparisons.
The default view, including all records regardless of grade classification. This provides the broadest statistical picture but mixes parts with very different design philosophies and intended radiation environments.
Commercial parts not specifically designed or qualified for radiation environments. COTS devices are manufactured using standard commercial processes optimized for cost, performance, and power — not radiation hardness. Their TID tolerance is incidental to the process technology and can vary dramatically between manufacturers, process nodes, and even production lots [2, 3].
COTS parts are of particular interest because they offer significant advantages in cost, availability, performance, and power consumption compared to radiation-hardened alternatives. Many modern space programs use COTS devices in radiation environments, accepting managed risk through shielding, redundancy, system-level mitigation, or acceptance testing of flight lots.
Parts specifically designed, manufactured, or qualified for radiation environments. This category includes:
Radiation-hardened (Rad-Hard) devices designed using Hardness-By-Design (HBD) techniques such as enclosed-layout transistors (ELT), guard rings, triple-well isolation, and special gate oxide processes [4]. These devices are intentionally engineered to withstand specified TID and SEE levels.
Space-grade parts that have been characterized and screened for space applications, potentially including lot-specific TID and SEE testing, extended temperature range qualification, and enhanced reliability screening.
QML-qualified (Qualified Manufacturers List) parts produced on production lines certified to MIL-PRF-38535 quality and reliability standards, with radiation lot acceptance testing (RLAT) performed on each production lot.
Mixing COTS and rad-hard devices in the same distribution can be misleading. The TID survival curve for "All Parts" will show a bimodal character: a population of COTS devices failing at lower TID levels and a population of rad-hard devices surviving to much higher levels. Filtering by grade allows the engineer to see the distribution relevant to their specific parts selection approach.
The year range slider filters the dataset by REDW publication year (1992–2025). This is important because the radiation hardness of electronic devices is strongly influenced by the semiconductor process technology, which has evolved dramatically over this period.
Over the 33-year span of REDW data, mainstream CMOS technology has scaled from roughly 800 nm (early 1990s) to 5 nm and below (2020s). This scaling has significant and sometimes counterintuitive effects on radiation hardness [2, 3, 5]:
Gate oxide thickness. As gate oxides have thinned with each technology node, TID-induced threshold voltage shifts have generally decreased. Devices manufactured on sub-100 nm nodes often show significantly better inherent TID hardness than older technologies, because thinner oxides trap less charge [2, 5].
Isolation oxides. While gate oxide TID effects have improved with scaling, isolation structures (STI, LOCOS) have not scaled as aggressively. TID effects in isolation oxides — including leakage current increases and parasitic transistor activation — have become the dominant TID failure mechanism in modern CMOS [3].
Supply voltage scaling. Reduced operating voltages in advanced nodes decrease noise margins, potentially making devices more susceptible to radiation-induced parametric shifts even if absolute threshold voltage changes are smaller.
SEL susceptibility. Technology scaling has mixed effects on SEL. Reduced feature sizes can decrease the sensitive volume for charge collection, but reduced well doping and thinner epitaxial layers can change the latchup holding conditions [6].
A device tested in a 1995 REDW paper was manufactured on a process technology (e.g., 500 nm CMOS) that is fundamentally different from the process used to manufacture the same part number today. Many long-lived commercial part numbers have undergone multiple die shrinks, foundry transfers, or process changes over their production lifetime. A TID result from 1995 has limited predictive value for a device purchased in 2025.
For screening purposes, limiting the year range to the most recent 10–15 years (approximately 2010–2025) provides data more representative of currently available process technologies. However, for rad-hard parts with controlled and stable manufacturing processes, older data may remain valid if the manufacturer confirms no process changes.
The dataset classifies devices into 17 categories based on function. Device-type chip filters on the TID and SEL tabs allow the user to view distributions for specific categories or compare across categories using the Statistics Table.
| Category | Includes |
|---|---|
| BJT | Bipolar junction transistors (NPN, PNP), discrete and arrays. Subject to ELDRS concerns at low dose rates [8, 9]. |
| Op-Amp | Operational amplifiers, instrumentation amplifiers, difference amplifiers. Both bipolar and CMOS input stages. |
| Comparator | Voltage comparators, window comparators. |
| Voltage Ref | Voltage references (bandgap, buried zener, etc.). Often bipolar-based and potentially ELDRS-susceptible. |
| Linear Reg | Linear voltage regulators (LDO, standard). Both positive and negative regulators. |
| Power Converter | DC-DC converters, switching regulators, charge pumps, power management ICs. |
| ADC | Analog-to-digital converters of all architectures (SAR, sigma-delta, pipeline, flash). |
| DAC | Digital-to-analog converters. |
| FPGA | Field-programmable gate arrays (SRAM-based, flash-based, antifuse). Includes SoC FPGAs. |
| MOSFET | Discrete power MOSFETs (N-channel, P-channel), including single-event gate rupture (SEGR) and single-event burnout (SEB) susceptibility. |
| Optocoupler | Optocouplers, optoisolators, LED-photodetector pairs. Known to be ELDRS-susceptible [8]. |
| SRAM | Static random-access memory. A primary SEU concern due to the sensitivity of bistable storage cells. |
| Flash | Flash memory (NOR, NAND), EEPROMs, and other non-volatile memories. |
| Logic/Interface | Digital logic ICs (gates, buffers, flip-flops, multiplexers), bus transceivers, level translators, interface ICs (RS-232, RS-485, LVDS, CAN). |
| MCU/Processor | Microcontrollers, microprocessors, DSPs, SoCs with integrated processing cores. |
| Sensor | Image sensors (CCD, CMOS), temperature sensors, Hall-effect sensors, accelerometers, other MEMS and sensing devices. |
| Other | Devices not fitting the above categories: oscillators, PLLs, clock generators, specialty analog, mixed-signal ICs, and miscellaneous components. |
The REDW distributions are most valuable as a screening and risk assessment tool during early mission design phases, before device-specific test data is available. They provide a statistical context for understanding what radiation performance is typical, what is exceptional, and what is unlikely.
When evaluating a candidate part for a radiation environment, the distributions answer the question: "For this type of device, what fraction of tested parts would meet my requirement?" This is useful for:
Feasibility assessment. If the mission requires 100 krad TID tolerance and the COTS Op-Amp distribution shows only 5% of devices surviving to 100 krad, a COTS Op-Amp is a high-risk choice and a rad-hard alternative should be considered. Conversely, if 80% of COTS logic devices survive to the required level, a COTS approach is more viable.
Technology trade studies. Comparing distributions across device categories helps identify which subsystems are likely to drive radiation design challenges. Categories with low TID survival at the mission dose level will need either rad-hard parts, additional shielding, or system-level mitigation.
Risk quantification. The percentile statistics provide a quantitative basis for risk statements: "Based on historical REDW data, the P10 TID hardness for COTS voltage references is X krad, meaning roughly 10% of tested devices in this category failed below X krad."
The percentiles can be mapped to approximate risk levels for an untested device randomly selected from the historical population:
| Percentile | Interpretation | Risk Posture |
|---|---|---|
| P10 | 90% of tested devices met or exceeded this level | Conservative / worst-case screening |
| Median (P50) | 50% of tested devices met or exceeded this level | Typical expectation |
| P90 | Only 10% of tested devices exceeded this level | Optimistic / best-case expectation |
For a complete assessment, the TID survival distribution should be compared against the mission total dose requirement (including radiation design margin, typically 2× per ECSS or MIL-STD-814). Similarly, SEL LET thresholds should be compared against the effective LET environment for the mission orbit and shielding configuration.
The TID threshold slider in the tool allows the user to set a reference level corresponding to the mission TID requirement. This highlights the fraction of devices in each category that would meet the requirement based on historical data, and filters the Parts List to show only devices that passed at or above that level.
The REDW distribution data is a powerful resource, but proper interpretation requires understanding its limitations. The following caveats should be considered when using this tool for engineering decisions.
As discussed in Sections 3 and 4, both TID and SEL data are predominantly right-censored. Reported pass levels are lower bounds on true failure thresholds. The observed distributions systematically underestimate the true hardness of the tested population. This bias is conservative (the true situation is better than shown), but it means the distributions should not be interpreted as representing actual failure thresholds.
REDW data is compiled from tests performed at different facilities, using different radiation sources, dose rates, bias conditions, and test methodologies over three decades. Key variations include:
TID dose rate. High dose rate testing (50–300 rad(Si)/s per MIL-STD-883 TM1019) may not capture ELDRS effects in bipolar technologies. Some REDW entries include low dose rate data, but most are high dose rate results [8, 9, 10].
Bias conditions. TID degradation is bias-dependent. Worst-case bias for CMOS is typically "all terminals grounded" or "VDD applied with inputs/outputs grounded," but specific bias configurations vary between test programs.
Pass/fail criteria. Different programs define "functional" differently. Some test only for parametric degradation against datasheet limits, others test for functional operation in the intended application circuit. The TID "pass level" can vary depending on which parameter or function drives the failure.
SEL test conditions. SEL testing varies in ion species and energy (affecting the maximum achievable LET), temperature, bias voltage, and the criteria used to define latchup (current increase threshold, destructive vs. non-destructive).
A single REDW entry typically represents one or a few devices from a single production lot. Commercial semiconductor manufacturing exhibits significant lot-to-lot variation in radiation hardness due to process parameter fluctuations (oxide thickness, doping profiles, contamination levels). A device from one lot may pass 100 krad while the same part number from a different lot fails at 20 krad [2, 4]. The REDW distributions average over this variation but cannot predict the performance of any specific lot.
The REDW distributions provide population-level statistics, not device-specific characterization. They are appropriate for:
• Early design phase parts screening and feasibility assessment
• Technology trade studies and risk comparisons
• Developing intuition about radiation hardness across device categories
• Identifying device categories that may require rad-hard alternatives
They are not appropriate as a substitute for:
• Device-specific TID or SEE characterization testing
• Flight lot acceptance testing (RLAT)
• Formal radiation hardness assurance qualification
• Worst-case analysis for mission-critical applications
The REDW dataset does not represent a random sample of all electronic devices. It represents devices that the radiation effects community chose to test. This introduces several forms of selection bias:
Interest bias. Devices are tested because they are candidates for space or military applications. Commodity consumer parts with no plausible space application are underrepresented.
Survivorship bias. Devices known to be radiation-soft from prior experience may not be retested and reported, while devices with promising radiation performance may be tested repeatedly across multiple lots and programs.
Publication bias. Programs that find interesting or useful results (very hard or very soft parts) may be more likely to publish than programs with unremarkable results.
Technology bias. The mix of device types tested has shifted over time with technology trends. FPGAs and advanced processors are more heavily represented in recent years, while discrete bipolar devices were more common in earlier publications.