Chapter 3 · Lesson 2 of 6 · about 10 minutes
Flowdown in both directions
A requirement runs from the mission intent down to a function, then to a circuit or a part, and the evidence has to run back up the same path. Requirements flow down; test and analysis results must come back up to close the loop.
Connect the requirement to real hardware.
Four levels
The chain has four levels: the mission need, the allocation to a subsystem or function, the criteria placed on a part, and the verification that closes each level. A link between two levels records a dependency, so a change at one level has to be checked at the levels it connects to.
Shared resources
Two channels that look independent on the block diagram can share a regulator, a clock, a reset line, a memory or a software service, and one event in the shared item can take both channels down. Allocate the budget according to what actually fails together in the hardware rather than to the block diagram.
Reconcile the top-down budget with the bottom-up response
A subsystem budget says what the design is allowed to do. The part data say what it can do. The two meet in the circuit: how the part is used, what workload it runs, what mitigation surrounds it and which operating modes it sees. Where they do not meet, the gap has to be closed through one of those, and the reconciliation should be written down.
Allocation. The event contributions of individual devices can be added up only when they are expressed in the same terms (the same interval, the same exposure case) and only when it is clear what each event does to the system. Dividing a probability budget equally among the parts is tempting, but it is only valid if the parts fail independently, and that should be checked first.
Work product. Each requirement should carry an ID, its parent, the hardware it applies to, the criterion, the conditions, an owner, the verification method, the evidence and a list of open items. With those fields filled in, a reviewer can trace it in either direction.
Exercise
Take one part-level radiation criterion and trace it up to a measurable consequence at the system level.
A list of parts that each passed a test does not, by itself, verify a system function; the trace is what connects the two.
Going deeper
Allocation without hidden dependence
The independent event contributions, the common causes and the coupling through recovery are three different things, and a budget that adds up the first while ignoring the other two misses the behavior that matters. A numerical budget is useful only when the way it splits the system keeps those effects visible.
Model limit. The allocation arithmetic assumes that events arrive independently as a Poisson process and that each recovery succeeds independently. It leaves out correlated events, common-cause failures, the state of the workload at the time of the event, and the ways protection circuits interact with one another.
Practical review. Ask what happens if the recovery itself resets a shared service, if several channels share a power rail, or if the event lands during startup or a maneuver. Those cases need their own verification.
Takeaway
Keep each requirement attached to its parent and to its evidence.
Check yourself
Two redundant units share one regulator. Why does the allocation have to say so?
Because a regulator event takes down both units at once: the function's rate is the regulator's rate plus the independent rates, not the product of two unit probabilities. An allocation that treats the units as independent overstates the availability of the function.
A requirement is allocated as a rate per device. What has to travel with it for the part engineer to work with it?
The event definition (what counts), the environment case the rate refers to (quiet, worst week, peak), the time at risk and the consequence that set the number. A bare rate cannot be tested against or traced back to the function.
A function's probability budget is divided equally among the parts on its board. What has to be true for that to be valid, and what should be checked first?
The division is valid only if the parts fail independently, and only when the contributions are expressed in the same terms (the same interval, the same exposure case) and it is clear what each event does to the system. Shared resources should be checked first: a regulator, a clock, a reset line, a memory or a software service can take several channels down with one event, so the budget should be allocated according to what actually fails together in the hardware rather than to the block diagram. Where the independence does not hold, the allocation arithmetic generally overstates the availability of the function.
Every part on a board has passed its radiation test, and the list of passed parts is offered as verification of the system function. Why is it not?
A list of parts that each passed a test does not, by itself, verify a system function; the trace is what connects the two. The subsystem budget says what the design is allowed to do and the part data say what it can do, and the two meet in the circuit: how the part is used, what workload it runs, what mitigation surrounds it and which operating modes it sees. The evidence has to come back up the same four levels (mission need, allocation, part criteria, verification), with each requirement carrying its parent, its criterion, its conditions, its verification method, its evidence and its open items.
The allocation arithmetic adds up independent Poisson contributions. Which behaviors does it leave out, and what review questions find them?
It assumes that events arrive independently and that each recovery succeeds independently, so it leaves out correlated events, common-cause failures, the state of the workload at the time of the event and the ways protection circuits interact with one another. The independent contributions, the common causes and the coupling through recovery are three different things, and a budget that adds up the first while ignoring the other two misses the behavior that matters. The review asks what happens if the recovery itself resets a shared service, if several channels share a power rail, or if the event lands during startup or a maneuver, and those cases need their own verification.