What is successive refinement of the UPF, and how does -update work?

Short Answer

Successive refinement means power intent grows in layers: the IP provider writes constraint UPF, the integrator adds configuration such as strategies and control signals, and implementation adds supplies and cell detail. Each later layer refines earlier commands with -update instead of rewriting them, so an IP's constraints travel unchanged into every chip that uses it.

Technical Explanation

• Constraint UPF: atomic domains, set_retention_elements (UPF) for state that must be kept, clamp values on ports, and legal power states without voltages.
• Configuration UPF: the integrator adds isolation, level-shifter and retention strategies with control signals for this chip's use of the IP.
• Implementation UPF: supply sets and their association to domain handles, switches, isolation and retention supplies, and voltages.
• Refinable commands: the first call names the object; later calls in the same scope repeat the name (and domain, for strategies), add -update and only new arguments.
• Rules: -update on the first call is an error, a later call without it is an error, and a conflicting value for the same argument is an error.
• Deferred detail: add_power_state (UPF) can define ON as FULL_ON with no voltage; ICC2 needs the value filled with -update before any action or check command.
• Legacy form (still accepted by ICC2/PT): create_pst (UPF) with add_pst_state (UPF) for states, and set_isolation_control (UPF) for the control signal.

# [UPF]  cop_constraint.upf (IP provider)
create_power_domain PD_COP -elements {.} -atomic -supply {primary}
set_retention_elements RL_COP -elements {u_regfile u_fsm}
set_port_attributes -elements {dbg_valid} -clamp_value 0
add_power_state PD_COP.primary -supply -state {ON -supply_expr {power == {FULL_ON}}}
# [UPF]  mychip_config.upf (integrator)
set_isolation ISO_COP_OUT -domain PD_COP -elements {dbg_valid} -clamp_value 0 -isolation_signal ise -isolation_sense high
set_isolation ISO_COP_OUT -domain PD_COP -update -elements {irq_out}
# [UPF]  mychip_impl.upf (implementation)
associate_supply_set SS_COP -handle PD_COP.primary
set_isolation ISO_COP_OUT -domain PD_COP -update -isolation_supply SS_AON
add_power_state PD_COP.primary -supply -update -state {ON -supply_expr {power == {FULL_ON 1.0}}}
# [ICC2]  icc2_shell
load_upf mychip_top.upf
report_power_domains

What To Check

• Each layer lives in its own file with one owner.
• Every refining call repeats the object name and domain and uses -update.
• No later layer contradicts an earlier value, such as a clamp value.
• Deferred items such as state voltages are filled before implementation checks run.

Healthy, Suspicious & Hard-stop Results

• Healthy (illustrative): All three layers load without errors, and report_power_domains shows PD_COP with SS_COP as its primary supply.
• Suspicious (illustrative): A power state is still FULL_ON with no voltage when implementation starts.
• Hard stop: load_upf errors on a set_isolation that repeats ISO_COP_OUT without -update, or with a different clamp value.

What The Interviewer Is Testing

• Can you name the three layers and who owns each?
• Do you know the -update rules and what a refinable argument is?

Follow-up Question & Model Response

"Can you change an atomic domain's elements during refinement?"

Candidate Model Response: No. IEEE 1801 makes it an error to use -elements or -exclude_elements with -update on an atomic power domain, and an error to add -atomic with -update. You can still add supplies and compose it with other domains, but you cannot split it. That is what atomic means: the IP owner says this logic must power up and down as one unit.