Chapter 4 · Lesson 4 of 6 · about 24 minutes
Total ionizing dose in MOS structures
Ionization can leave trapped charge in the insulators and create defects at the interfaces. As the dose accumulates, the electrical behavior of the device shifts. A TID requirement should protect the parameters and functions the application depends on, including the parameter changes that come before any outright failure.
The space-rha.com TID model: oxide trapped charge and interface traps under bias, dose rate, temperature and annealing, with the transfer curve and the time history.
An original space-rha.com laboratory, embedded unchanged with its own controls; its onset values, gains and depths are teaching assumptions, not limits for flight hardware.
Accumulated ionization changes the conditions inside the device.
A small threshold shift can hide competing damage.
Positive oxide-trapped charge and interface traps can push the threshold in opposite directions under some bias and device conditions. The leakage, mobility, subthreshold slope and noise can degrade even while the net threshold movement stays modest. The material the dose is quoted in, the bias, the temperature and the elapsed time are all part of any comparison between two results.
Measure several parameters, at stated times, so that the ones that moved are not hidden behind the one that did not. A circuit should not be qualified from a single threshold-voltage measurement on one transistor.
Evidence to retain. Record the dose and the material it is quoted in, the source spectrum, the dose rate, the bias, the temperature, the timing of each measurement, any anneal and the parameter limits that were applied.
Mission context. A timing chain can drift outside its allocated margin while every device in it still passes a simple power-on check.
The isolation oxide can matter more than the gate oxide.
A thin modern gate dielectric can close off one TID pathway while the thicker isolation regions remain susceptible. Leakage through those regions can raise the standby power, shift operating points or corrupt weak signals. A smaller technology node is therefore not a guarantee of hardness.
A parasitic leakage path can break the application budget.
Charge trapped in the isolation oxide can turn on conduction along a transistor edge or between devices, depending on the geometry and the bias. In a precision or low-power application, that leakage can be unacceptable at a dose far below the one that causes a digital functional failure. Layout techniques can help, but their benefit has to be shown for the process and the endpoint in question.
Measure the off-state and standby currents, the analog input leakage and whatever other low-current behavior the application depends on, across the supply and temperature range it will see.
Evidence to retain. Check the geometry, process, layout and bias of the flight part against those of the parts behind the supporting data. Record how sensitive the parametric measurement was and how the result varied with temperature.
Mission context. A battery-limited sensor can exceed its standby allocation while its digital logic still operates correctly.
The circuit limit is the endpoint that matters.
Changes in threshold, leakage, offset, gain, noise and timing are intermediate observables. The application may have a much tighter limit than the manufacturer’s full operating specification. The measured drift has to be tied to the function it affects and to the margin that function was allocated.
The operating point can turn a mild shift into a large effect.
Headroom, feedback, load, reference accuracy and the conditions at startup determine what a parametric shift does to the circuit. Two shifts can compensate at one operating condition and reinforce each other at another. A board test can reveal such interactions, but limited access to internal nodes and differences in source and dosimetry limit what it can say about each part.
Measure the operating corners the application requires, including its degraded configurations, and trace each limit back to a circuit analysis or to an endpoint that was observed in the application.
Evidence to retain. Keep the values before and after exposure, the correlations between parameters, the corner conditions, the reason the source applies, the identity of each sample and the reasoning behind each application limit.
Mission context. A converter can meet its broad output-voltage specification while exceeding a precision instrument’s reference-error allocation.
The response depends on time as well as dose.
Damage builds and recovers during the irradiation, between measurements, and later in storage or in operation. Bias and temperature influence both the buildup and the recovery. A single final measurement can miss a worst case that occurred part way through, and it can miss a parameter that later reverses direction.
The test history is part of the evidence.
The bias during irradiation sets the electric fields in the oxides and the state of the circuit. A delay before measurement can hide a shift that has already partly recovered. Oxide-trapped charge may anneal while the interface traps persist or keep growing, so a parameter can rebound past its starting point or a different parameter can take over as the limiting one. More annealing therefore does not necessarily mean more margin.
The record should keep the dose steps, the times at which the exposure was interrupted, the bias during measurement, the temperature and the sequence of post-irradiation measurements. Where the application has them, powered, unpowered and switched modes should each be assessed.
Evidence to retain. Record elapsed time alongside dose, and follow the parameter that matters most through the whole post-irradiation history the requirement calls for.
Mission context. A cold cruise followed by warm powered operation takes the part through a different response history than a laboratory exposure that was warm throughout.
Find the vulnerable material in each technology.
An SOI buried oxide, a charge-storage memory, a high-k gate stack, a bipolar surface, an optical material and a sensor each respond to dose in their own way. Before data from one technology are applied to another, identify what in the new technology stores charge, transports carriers or carries the signal.
The same dose can drive different endpoints.
In a floating-gate memory the read window may shift. In a buried oxide the back channel may change. A high-k stack may hold a different population of traps. An optical link may develop changes in absorption or emission. The material the dose is quoted in and how the source deposits its energy remain part of any comparison.
Test the state that matters for the device, such as the programmed and erased distributions, retention, read margin, the optical wavelength or the back-gate bias, and qualify the exact endpoint the application depends on.
Evidence to retain. Record the process and stack, the memory state or the wavelength, the material the dose is quoted in, the bias and the timing. Look for data on that specific technology and for a model that has been validated against it.
Mission context. A photonic receiver and its digital controller see the same exposure, but they fail through different materials and against different performance limits.
A dose test is evidence under stated conditions.
A krad number on its own says little. It needs the material, the source, the geometry, the dose rate, the bias, the temperature, the number of samples and the definition of failure that went with it. A test result supports a claim about the application only when those conditions represent or bound the application, and the report says why.
The same dose label does not guarantee the same damage.
Co-60, X-rays, electrons, protons and heavy ions deposit their energy differently, with different recombination, different dose enhancement and, for the particles, displacement damage at the same time. Board shielding and high-Z packaging can change the dose at the die. Samples and lots can respond differently even under the same commercial part number.
Choose the experiment to answer the question that was asked. A board test can show how the application behaves, but it needs an appropriate source and dosimetry, coverage of the operating conditions, and a reason to believe the parts on that board represent the population that will fly.
Evidence to retain. Record the dosimetry and its uncertainty, the conversion between materials, the exact part and lot, how many samples were tested, the endpoints, the sequence of exposures and the range over which any model applies.
Mission context. A board test can demonstrate a useful system response and still say nothing about a lot that was not on the board, or about a part inside the board that the shielding protected.
A question to carry through the course
When, if ever, can the dose applied to a part during a proton SEE test be used as a proxy for a TID test?
Things to weigh here:
- The dose rate: a run of 10¹⁰ protons per cm² at 200 MeV deposits about 0.7 krad(Si) in minutes, a few rad(Si) per second, which is below the 50 to 300 rad(Si)/s condition of MIL-STD-883 TM1019 and far above its low-dose-rate condition.
- The bias: the part sat at the operating, flight-like bias of the SEE test, which may not be the bias that traps the most charge.
- The measurements: which parameters, if any, were measured, when they were measured relative to the exposure, and whether any anneal followed.
Going deeper
Traps alter electrostatics and carrier exchange.
Ionization creates carriers in the oxide, and field-dependent recombination and transport decide how many survive to be trapped. Oxide-trapped charge changes the potential. Interface states exchange charge with the silicon according to their energy, their capture and emission kinetics and their occupancy, which changes both the charge balance and the transport in the channel.
The planar ΔVot = −Qeff/Cox relation is electrostatics only. The sign of the interface charge depends on occupancy, and the yield of traps per unit dose differs from one oxide to another.
Three-dimensional sidewall electrostatics controls edge conduction.
Trapped charge and interface states modify the potential at the corners and sidewalls of the isolation trench. Whether a parasitic inversion path forms depends on the channel geometry, the doping and how much control the gate has there. The mobility and the interface-state response can rise and fall with dose and annealing rather than moving in one direction.
A planar capacitor illustrates the charge coupling but cannot solve a trench corner. A quantitative leakage claim needs calibrated 2D or 3D electrostatics and transport.
Sensitivity analysis identifies the controlling parameters.
A compact model can translate changes in threshold, mobility, leakage or bipolar gain into circuit behavior, provided the radiation-induced shifts and the correlations between them are backed by data. A SPICE model fitted before irradiation does not by itself predict the irradiated device or the spread across a lot.
Propagate the parameter changes together, with their correlations and their recovery history. When two mechanisms evolve on different timescales, a net result that looks like cancellation should be treated with caution.
A spectrum of traps gives a spectrum of times.
Thermal emission, tunneling, hole transport and the reactions at the interface each have their own activation energy and field dependence. A single exponential is a deliberately reduced model; real distributions of time constants can extend over many decades. The net threshold motion can reverse when the opposing contributions decay at different rates.
A single measured time constant should not be extrapolated to other temperatures, fields, trap populations or mission durations. The mechanism and the history that apply to the part need their own evidence.
Trap location and band structure set the coupling.
Traps near different interfaces have different electrostatic weight and different capture and emission times. Charge-storage devices add tunneling barriers and retention paths. In optical materials, color centers and defect transitions change the absorption or the luminescence, and the planar silicon capacitor model does not describe any of that.
Each technology needs its own physical model. The ionization that starts the process is the same in all of them, but the response that follows and the dosimetry conversion are specific to each.
Coupled mechanisms can alter the next response.
Accumulated ionizing or displacement damage can change the leakage, the gain, the charge collection and even the SEE sensitivity. The change is not necessarily for the worse, and a reduction in one endpoint does not establish that the part is hard in general. The order of the tests and the exposure accumulated along the way can therefore bias an SEE measurement or a comparison between dose results.
Where it matters, model the charge yield and the local deposition for the specific source. Keep the TID and displacement damage accumulated by each device and its condition after the test. The dose accumulated during a heavy-ion test is not, by itself, a TID qualification.
The common mistake
The threshold-contribution plot in the laboratory does not predict STI leakage quantitatively.
The electrostatic example gives no calibrated circuit transfer, no lot distribution and no margin policy.
The normalized kinetic model has one assumed time constant and includes neither rebound nor hydrogen chemistry.
The SiO₂ model is deliberately limited: it does not apply to optical color-center spectra or to many of the alternative dielectrics.
A teaching curve of the electrostatics supplies none of the sample statistics or the source-equivalence evidence a qualification needs.
Takeaway
- A dose requirement should trace to the circuit performance that has to survive, not only to a nominal dose number.
- The technology review should include the isolation structures and the low-current endpoints.
- Judge dose tolerance against the application limits and the uncertainty in them.
- Control the whole history from exposure to measurement, and report it.
- Before transferring data from another technology, match the vulnerable structure and the observable response.
- Source equivalence, inference from the samples to the population, and coverage of the application are three separate arguments; keep them separate.
Check yourself
A CMOS part with a very thin gate dielectric is described as inherently tolerant to dose because the gate oxide is too thin to trap much charge. What does that argument leave out?
A thin modern gate dielectric can close off one TID pathway while the thicker isolation regions remain susceptible. Charge trapped in the isolation oxide can turn on conduction along a transistor edge or between devices, which raises the standby power, shifts operating points or corrupts weak signals, and in a precision or low-power application that leakage can be unacceptable at a dose far below the one that causes a digital functional failure. A smaller technology node is therefore not a guarantee of hardness; the off-state and standby currents and the analog input leakage have to be measured across the supply and temperature range the application will see, and layout techniques are credited only once their benefit has been shown for the process and the endpoint.
A TID report shows the threshold voltage of a test transistor moving only slightly up to the required dose, and the circuit is declared qualified on that basis. Why is that insufficient?
Positive oxide-trapped charge and interface traps can push the threshold in opposite directions under some bias and device conditions, so a modest net shift can hide competing damage: the leakage, mobility, subthreshold slope and noise can degrade while the threshold barely moves. The two contributions also evolve on different timescales, so a result that looks like cancellation should be treated with caution. Several parameters should be measured at stated times, and a circuit should not be qualified from a single threshold-voltage measurement on one transistor; the endpoint is the circuit performance that has to survive and the margin it was allocated, which may be much tighter than the manufacturer's full specification.
A part is irradiated to the required dose, left unbiased for a time before it is measured, and passes. A colleague says the delay was conservative because annealing only helps. What is wrong?
A delay before measurement can hide a shift that has already partly recovered, and more annealing does not necessarily mean more margin. Oxide-trapped charge may anneal while the interface traps persist or keep growing, so a parameter can rebound past its starting point or a different parameter can take over as the limiting one, and a single final measurement can miss a worst case that occurred part way through. The record should keep the dose steps, the interruptions, the bias during irradiation and during measurement, the temperature and the sequence of post-irradiation measurements, and where the application has them, the powered, unpowered and switched modes should each be assessed.
A board was irradiated with Co-60 to the mission dose and operated correctly throughout. The project wants to close the TID question for every part on it. Which three arguments does the course say have to be made separately, and what does the board test not show?
Source equivalence, inference from the samples to the population, and coverage of the application. A board test can demonstrate a useful system response, but it needs an appropriate source and dosimetry, coverage of the operating conditions and a reason to believe the parts on that board represent the population that will fly; it says nothing about a lot that was not on the board or about a part inside the board that the shielding protected. Limited access to internal nodes also means that a drifting parameter can be compensated by the surrounding circuit until it runs out of range (lesson 5.4), so the part-level endpoints the application depends on may not have been observed at all.
A proton SEE run of 10¹⁰ protons per cm² at 200 MeV deposited about 0.7 krad(Si) in a few minutes. Compare that dose rate with the conditions of MIL-STD-883 TM1019 and say what the comparison does and does not settle about using the run as dose evidence.
A few rad(Si) per second sits below the 50 to 300 rad(Si)/s standard condition of TM1019 and far above its low-dose-rate condition of 0.01 rad(Si)/s or less (lesson 6.2), so the run matches neither standard window, and for a bipolar or BiCMOS linear part whose response depends on the dose rate (lesson 4.5) that alone leaves the result without a reference condition. The dose rate is only one of the factors: the bias during the run was the flight-like operating bias, which may not be the one that traps the most charge, and whether any parameter was measured, when it was measured relative to the exposure and whether an anneal followed decide what the run can show at all. The course leaves the question open deliberately and carries it to lesson 5.4.
Further reading
- NASA Avionics RHA Guidelines, NASA/TM-20210018053 (2021), §5 TID
- G. R. Allen, "Predicting On-Orbit Performance of Devices," RADECS 2025 Short Course, Slides 4–5, 10–11, 12–22
- Oldham & McLean (2003), Total Ionizing Dose Effects in MOS Oxides and Devices, Figs. 1–3
- IEEE NSREC 2024 Short Course, "Short Course"
- IEEE NSREC 2020 Short Course, "Short Course"
- Schwank et al. (2008), Radiation Effects in MOS Oxides
- Space-RHA visual explanations
- G. R. Allen, "Predicting On-Orbit Performance of Devices," RADECS 2025 Short Course, Slides 2–10, 25–30
- Space-RHA: ELDRS tool
- IEEE NSREC 2013 Short Course, "Short Course"