space-rha.comShort courseChapter 5 · BOM scrub and parts review

Chapter 5 · Lesson 2 of 6 · about 12 minutes

Function first, then three radiation questions

One regulator can feed the processor, the memory and the recovery controller. One ADC can serve a single housekeeping measurement or close a critical control loop. Trace each part to its function and its recovery path before deciding which radiation concern to work first.

One small part can stop several functions.

A regulator may feed the processor, the memory and the controller that is supposed to recover them. An ADC may affect one housekeeping measurement, or it may close a control loop the mission depends on. Before you decide which radiation concern to work first, trace the part to the function it supports and to the path by which that function would be recovered.

The same part with the same exposure can have very different consequences depending on where it sits in the circuit.

Ask the designer about the actual circuit.

Sit down with the designer and record the rails, bias, load, duty cycle, temperature, operating modes, interfaces, timing and the parameter limits the circuit can tolerate. Map what is shared: power, clocks, reset and fault detection. Then check whether the protection and recovery circuits stay powered, and stay visible in telemetry, during the failure they are meant to handle.

A spare channel does not help if both channels share the rail or the reset that failed.

Ask three separate radiation questions.

Total ionizing dose can shift parameters as exposure accumulates. Displacement damage can reduce carrier lifetime and optical response. A single event can interrupt a function, corrupt data or destroy the part. The technology and the function tell you which of these endpoints to investigate for a given part, and a passed dose test says nothing about the single-event question.

Use the technology to focus the first pass.

The technology points to where to look first: TID and SETs in analog parts, SEU and SEFI and possible destructive modes in complex logic, bias-dependent destructive SEE in power devices, and DD in minority-carrier and optical chains. These are starting points for the investigation and do not by themselves assign a susceptibility; a part can still surprise you outside its category. Conventional passives are usually a lower priority for these mechanisms, but their construction, their use in the circuit and their other hazards still get reviewed.

When you decide an effect is out of scope for a part, write down why. A part with no data is a different finding from a part whose mechanism has been ruled out.

The first pass, by technology

Before you search for any data, sort the parts list by technology. Any part with CMOS in the die can latch up. Any part with memory elements can upset. Any part with control logic or registers can stop in a state it does not leave on its own. Analog paths, clocks and regulators produce transients, and what those transients do is decided by the circuit that receives them. Power devices under bias can burn out or rupture. Minority-carrier and optical devices lose performance to displacement damage. Every active part accumulates dose. None of these is a finding about the part; each is a question the part raises. The chart below walks one BOM line through the questions, the tabbed set after it does the same for the common device families, and lesson 5.4 shows how a question is then closed with evidence. Four outcomes are possible for each question: closed by evidence that applies, an action (a targeted test, an analysis, supplier data, a design change or a replacement), bounded by the design, or accepted by the project as a documented residual with its scope stated.

Flow chart

The first-pass screen for one BOM line

Every question is asked of every part; a yes opens the box beneath it, and the dose question applies to every active part. A yes means evidence is needed; it does not mean the part is susceptible. Walk one of the teaching parts through the chart, or select a box to see what the question means.

  1. A 1 Gbit CMOS DRAM with its controller, one manufacturer, one orderable number, one die revision. Every question opens: the CMOS die raises latchup, the array and the mode registers raise upsets with multiplicity, the refresh controller raises interrupts, and dose applies as it does to every active part. Four questions, each handed to a later chapter.

  2. A bipolar linear process without CMOS, so the latchup question is not raised; no memory elements and no control logic. The analog path raises transients, judged at the comparator it feeds, and the bipolar gain raises displacement damage, generally closed by analysis at this fluence. The dose question, with ELDRS, is the one that stays open into chapter 11.

  3. An LED, a photodetector and a bipolar receiver amplifier in one package, so the construction is part of the identity. The receiver is an analog path, where a transient is a false edge on the link, and the LED and detector raise displacement damage, which is the dominant question for this part.

  4. Not an active semiconductor or an optical device, so it sits outside the three questions, and the record states the reason. A resistor in a high-voltage divider or in an optical path would call for a look at its own construction.

Which radiation questions does a part raise? A first pass by technology no yes yes yes yes yes yes yes yes yes BOM line with its exact identity Active semiconductor oroptical device? Passive: generally outside thesequestions; record the reason and anyconstruction exception Any CMOS or BiCMOS inthe die? SEL question: latchup at maximumvoltage and temperature(destructive-class consequence) Memory elements:registers, SRAM, DRAM,flash, configuration? SEU and MBU question: upsets,multiplicity, correctability Control logic, statemachines, an embeddedcontroller? SEFI question: interrupt modes andthe recovery rung that clears each Analog path, clock,comparator, regulator? SET question: amplitude andduration against the receiverthreshold; propagation on a rail Power device underdrain or collector bias(Si, SiC, GaN, IGBT)? SEB and SEGR question: bias-dependent, destructive; safeoperating area by derating Minority-carrier oroptical (LED, detector,optocoupler, imager,bipolar gain)? DD question: output, gain, darkcurrent or CTR against the missionproton (and electron) fluence TID question for every active part: MOS leakage and threshold; bipolar gain and biascurrents with ELDRS; the parameters the circuit needs, at the mission dose with the designmargin Each question ends in onedisposition: closed by evidence,action needed, bounded by design, oraccepted

Flow chart

The questions in order, by device family

Select a family. Each chart gives the order in which the questions are generally worked for that family: the destructive question first where there is one, then the recoverable single-event questions, then the cumulative ones, ending at the consequence for the circuit, which chapters 9 and 10 compute.

  1. An SRAM-based FPGA with an embedded processor. No high-current events to the highest LET tested at maximum voltage and 125 °C, so a bound is stated with its fluence and conditions rather than immunity. Interrupts of the configuration controller and the processor are found at the flight workload, configuration upsets are the SEU question, with TMR and scrubbing verified on the bitstream, and the core current is checked at the mission dose.

  2. Seven LETs and two proton energies, with 1 to 4 % of events multi-cell and clusters up to 3 bits; refresh and mode-register interrupts cleared by reset, one of which needs a power cycle; repeated errors at one address logged as retention changes; refresh and standby current at 30 krad with margin. The interleaving, the scrub interval and the SEFI rate go to chapters 9 and 10.

  3. A bipolar linear process: the dose question is TID with ELDRS, the high- and low-dose-rate legs of lesson 6.2 with the worst-case bias. Transients of up to 1 V and 2 µs at the output are mitigated by a 20 µs low-pass filter at the comparator, and there is no digital control inside. The end-of-life offset error and the rate of consequential transients are the outputs.

  4. A CMOS process: high-current events above LET 21 at 125 °C and maximum input, and leakage under dose. Output transients reach every load on the rail, so the loads' brown-out thresholds and hold-up decide the consequence, and the sequencing controller with its configuration registers raises the SEFI question.

  5. A 200 V silicon vertical MOSFET switching an 80 V rail. The drain voltage is stepped at LET 37 with the gate at 0 V, with leakage checked after each step; the highest passing voltage is 150 V, 75 % of it gives 112 V, and the application stays below 100 V. Burnout and gate rupture are argued by avoidance, with no rate claimed; the gate driver is reviewed under the analog and digital flows.

  6. A GaAs LED with a silicon detector and amplifier. The current transfer ratio at the flight drive current is set against the five-year proton fluence of a sun-synchronous orbit: on the teaching curve it falls to 83 % of its initial value at five years, and the receiver needs 0.5 of 1.2. The LED material's own damage curve, not silicon NIEL, converts the spectrum.

Simplified teaching flows. Every family has exceptions (a CMOS part on a latchup-free process, a bipolar part without ELDRS, an optocoupler with a CMOS receiver). The charts put the questions in order; the evidence answers them.

Processors, FPGAs and ASICs: latchup first because its consequence is destructive-class, then interrupts and upsets, then dose; the configuration memory of an SRAM-based FPGA is a memory for the SEU question.
Digital CMOS: processors, FPGAs, ASICs no yes Digital CMOS part (processor, FPGA,ASIC) SEL screen at maximumvoltage, 125 °C, highLET with range at thewells: events? No events: state the bound (3 ÷live fluence at 95 %) with theconditions; keep as a claim Events: threshold and σ(LET);current limiter and recoveryverified with the rail capacitance;latent damage inspected SEFI: which modesinterrupt, and whichrung clears each(retry, reset, powercycle)? SEU: registers, caches,configuration (FPGA bitstream);multiplicity; mitigation by EDAC,scrubbing, TMR, configurationscrub; watchdog coverage TID: standby and active current,timing at end of life, at themission dose with margin; lot basis Rates per mode (chapter 9);consequence at the function;residual to the project
SRAM, DRAM and NAND flash: the upset cross section with its multiplicity feeds the code and scrub design of chapter 10; the technology decides the cumulative endpoints.
Memories: SRAM, DRAM, NAND flash DRAM flash SRAM Memory part SEU and MBU: cross section per bitagainst LET and proton energy, withpattern, mode and multiplicity; mapthe cluster size to the wordinterleaving SEFI: controller and moderegisters; flash charge pumpsduring program and erase; the rungthat clears each Technology? DRAM: stuck bits and retentionchanges that the scrubber reportsas repeats; refresh under TID Flash: data retention and programor erase failure after TID;cumulative write history SRAM: standby current and retentionvoltage under TID; configurationSRAM in FPGAs TID endpoints at the mission dosewith margin; lot basis for theflight parts Scrub interval and code sized fromthe rate and the clusterdistribution (chapter 10); SEFI rateto the availability budget
Amplifiers, comparators, references, regulators and converters: the process decides the dose question (ELDRS for bipolar, latchup and leakage for CMOS); transients are judged at the receiver; digital control inside a converter raises the SEFI question.
Analog and mixed-signal: amplifiers, comparators, references, regulators, converters bipolar CMOS, BiCMOS yes no Analog or mixed-signal part Process? Bipolar or BiCMOS linear: TID withELDRS; low-dose-rate test or avalidated factor; bias duringirradiation chosen from the worstcase CMOS: SEL question first (maximumvoltage, 125 °C); TID leakage andoffset drift SET: amplitude and durationdistribution from the beam, mappedthrough the receiver threshold andtiming; a regulator transientpropagates to every load Mitigation at the receiver:filtering, time qualification,range checks, voting; verify theslowed wanted signal Digital control inside(ADC, DAC, PLL, digitalpotentiometer)? SEFI question as for digital CMOS;registers that hold configuration End-of-life circuit analysis withthe degraded parameters; SET rate ofconsequential transients
Vertical MOSFETs, IGBTs, SiC and GaN transistors and high-voltage diodes: burnout and gate rupture are argued by avoidance (a derated safe operating area), not by a rate; the gate driver is reviewed under the analog and digital flows.
Power devices: vertical MOSFETs, IGBTs, SiC and GaN transistors, high-voltage diodes Si SiC, GaN IGBT, diode Power device under drain, collector orreverse bias in flight SEB and SEGR: step the drain (orcollector) voltage at a stated gatebias and a reference LET with rangeinto the epitaxial layer and body;leakage after each step; post-irradiation gate stress Safe operating area: a deratedfraction of the highest passingvoltage (a common convention is75 % of V_SEE at LET 37, gatecondition stated) Technology? Si MOSFET: threshold shift andleakage under TID with the flightgate bias; body diode under DD athigh fluence SiC and GaN: generally tested thesame way, with leakage degradationas the endpoint; the angulardependence is part-specific IGBT and diodes: burnout underbias; forward drop and leakageunder DD and TID Gate driver and controller: SET andSEFI (a spurious turn-on is thedestructive path); see the analogand digital flows Operating point inside the deratedregion; rate not claimed where themodel is uncertain; avoidance arguedinstead
LEDs, lasers, detectors, optocouplers, imagers and bipolar discretes: displacement damage first, with the endpoint the circuit uses and the material's own damage curve; the surrounding circuitry carries the dose and single-event questions.
Optoelectronic and minority-carrier devices optocoupler detector, imager emitter LED, laser diode, photodiode,optocoupler, imager, solar cell, bipolardiscrete DD question first: the endpoint thecircuit uses (light output, CTR atthe flight drive, dark current,responsivity, gain) against themission proton fluence, convertedthrough NIEL for the devicematerial Device? Optocoupler: LED, detector andamplifier degrade together; CTR atthe flight current and temperature;receiver threshold Detector and imager: dark currentand hot pixels (DD); readout logicSEU and SEFI; SET in the signalchain LED and laser: output versusfluence; drive margin; injectionannealing credited only withevidence TID: the CMOS or bipolar circuitryaround the optical element (imagerreadout, optocoupler amplifier) atthe mission dose with margin Material and wavelength: a siliconNIEL curve does not transferunchanged to GaAs, InGaAs or othermaterials; check at a second energy End-of-life margin of the link ormeasurement; drive, calibration orredundancy as mitigation

Exercise on the demonstration board

On the demonstration board, select U1 and U5 and compare what each one feeds and what its proposed recovery depends on. Then switch the evidence view between total dose, displacement damage and single events for one part and notice how the question changes with each view. Finally select U4, the optocoupler, and treat the LED, the receiver and the circuit threshold as one signal chain rather than one part.

Going deeper

Keep failure paths and dependencies visible.

Translate each part-level failure mode into the circuit and system state it produces: an output that drifts, a transient that the next stage captures, a bus held in contention, corrupted memory, a functional interrupt or a destructive current. For each one, follow where it propagates, whether anything detects it, whether it can sit dormant, what else shares the cause and when in the mission it would matter most. A block diagram shows the intended independence and recovery paths. It generally does not show the shared rail or the shared reset that defeats them, so do not infer either from the diagram alone.

SEECA and the circuit and fault analyses support this mapping. A component upset rate becomes a mission-loss rate only after it has been carried through the function and the recovery path.

Let the structure determine the mechanism boundary.

A technology name is a starting hypothesis about the device, and the device structure decides which mechanisms apply. Charge collection, parasitic paths, the oxide and interface response, minority-carrier lifetime and high-field regions each govern different effects, and a hybrid can contain several technologies in one package. Before you exclude a mechanism for the whole part, check the dose history, the material, the bias, the temperature and the endpoint that the exclusion depends on.

The mechanisms chapter gives the device models. This chapter is where you record whether that model applies to the part on this BOM line.

The common mistake

Calling a part payload or redundant does not make its failure tolerable. Trace the failure path before you accept the label, and do not exclude an effect on the strength of a part-number prefix.

Takeaway

  • Criticality comes from the function and the failure path, so trace both before ranking a part.
  • TID, DD and SEE are three separate questions with three separate evidence trails.

Check yourself

A regulator feeds the processor, the memory and the recovery controller, but the board has a redundant channel, so the designer rates the regulator as low criticality. What should the review check before accepting that?

Criticality comes from the function and the failure path, and a spare channel does not help if both channels share the rail or the reset that failed. Map what is shared (power, clocks, reset and fault detection) and check whether the protection and recovery circuits stay powered, and stay visible in telemetry, during the failure they are meant to handle. A block diagram shows the intended independence and generally does not show the shared rail or reset that defeats it, so neither should be inferred from the diagram alone; calling a part payload or redundant does not make its failure tolerable.

A passed TID test is on file for an ADC that closes a control loop, and the designer argues that the radiation question for that part is closed. What is missing?

TID, DD and SEE are three separate questions with three separate evidence trails, and a passed dose test says nothing about the single-event question. A mixed-signal converter produces transients that are judged at the receiving function, can deliver incorrect codes, and holds its configuration in registers, which raises the SEFI question of the digital flow; if the die is CMOS, the latchup question is open as well. U5 on the demonstration board shows the gap: DC parameter data exist, but the converter test covers one sample rate, the reference and SEFI coverage is incomplete, and the measured code errors still have to be mapped to the control loop rather than reported as a part-level count.

Walk U3, the bipolar amplifier, through the first-pass screen. Which questions does the chart not raise for it, and why is the dose question the one that stays open longest?

A bipolar linear process without CMOS, so the latchup question is not raised (a BiCMOS variant would raise it), and no memory elements and no control logic, so there is no SEU or SEFI question. The analog path raises transients, judged at the comparator it feeds (transients of up to 1 V and 2 µs against a 0.25 V receiving threshold, mitigated by a 20 µs low-pass filter, lesson 10.3), and the bipolar gain raises displacement damage, generally closed by analysis at this fluence. The dose question is TID with ELDRS, whose low-dose-rate leg takes weeks (lesson 6.2) and whose answer is lot-specific, so it is the one that stays open into chapter 11.

The first-pass screen opens a question for almost every part: any CMOS die can latch up, any memory element can upset, any controller can stop in a state it does not leave. Why does the course insist that a yes is not a finding about the part?

A yes means evidence is needed; it does not mean the part is susceptible. Each question the technology raises ends in one of four dispositions: closed by evidence that applies, an action (a targeted test, an analysis, supplier data, a design change or a replacement), bounded by the design, or accepted by the project as a documented residual with its scope stated. The converse also holds: when an effect is ruled out of scope, write down why, because a part with no data is a different finding from a part whose mechanism has been ruled out. Every family has exceptions (a CMOS part on a latchup-free process, a bipolar part without ELDRS, an optocoupler with a CMOS receiver), and the evidence, not the chart, answers the question.

U4, the optocoupler, is one line on the BOM. Why does the course ask you to treat the LED, the receiver and the circuit threshold as one signal chain, and which question dominates?

An optocoupler is an LED, a photodetector and a receiver amplifier in one package, so the construction is part of the identity: the receiver is an analog path on which a transient is a false edge on the link, and the LED and the detector raise displacement damage, which is the dominant question for this part. The endpoint is the current transfer ratio at the flight drive current against the mission proton fluence, converted through the damage curve of the LED material (GaAs here) rather than silicon NIEL, and judged against what the receiver needs: on the teaching curve the CTR falls to 83 % of its initial value at five years and the receiver needs 0.5 of the initial 1.2 (lesson 10.4). A CMOS receiver would add the latchup question.

Further reading