Skip to DFT, physical design, and signoff questions

ASIC implementation interview preparation

Practice the implementation questions first, then refresh the scan, floorplan, timing, power, and closure reasoning behind strong backend interview answers.

Question indexDFT, physical design, and signoff questions.

Filter the compact directory and open the exact selected question in the practice bank. Premium prompts remain access-gated.

Topic
Access

40 of 40 practice questions shown.

  1. Q0987Order the digital ASIC implementation flowFlow & signoff · EasySequence the digital ASIC flow and name each handoff artifact.Free
  2. Q0319Order an RTL-to-GDS flowFlow & signoff · EasyOrder RTL-to-GDS stages by what each consumes, changes, and proves.Free
  3. Q0973Match a physical-signoff failure to its fixFlow & signoff · MediumMatch timing, IR/EM, DRC, LVS, and antenna failures to root-cause fixes.Premium
  4. Q0803Scan shift versus captureScan & ATPG · EasySeparate scan shift from functional capture mode and clocking.Premium
  5. Q0277Generate a scan shift-capture sequenceScan & ATPG · MediumConstruct a legal load, capture, and unload scan timeline.Premium
  6. Q0145Eight-bit scan-chain RTLScan & ATPG · MediumCode an 8-bit scan chain with correct scan-enable behavior.Free
  7. Q0964Find vectors that expose internal stuck-at faultsScan & ATPG · MediumActivate and propagate internal stuck-at faults to observable outputs.Free
  8. Q0565Detect a transition-delay fault at speedScan & ATPG · MediumChoose launch and capture edges that expose transition-delay faults.Premium
  9. Q0359Model a four-bit MISR signatureScan & ATPG · MediumCompute a four-bit MISR recurrence and expected signature.Premium
  10. Q0182Contain unknowns before scan compactionScan & ATPG · MediumContain unknown responses before they poison scan compaction.Free
  11. Q0427Drive a complete March C-minus memory testMemory & access · HardApply the complete March C− address and operation sequence.Premium
  12. Q0160Architect scan, MBIST, and JTAG accessMemory & access · HardPartition scan, MBIST, JTAG, and test-controller ownership.Premium
  13. Q0694Four-bit JTAG instruction registerMemory & access · MediumImplement JTAG instruction capture, shift, update, and decode.Free
  14. Q0075Choose a pattern for an adjacent-row disturb faultMemory & access · EasySelect an operation sequence that exposes adjacent-row disturbance.Premium
  15. Q0894Choose the operation that exposes a stuck-at-zero SRAM cellMemory & access · EasyChoose the read/write operation that reveals a stuck-at-zero cell.Free
  16. Q1056Count unique detected faultsMemory & access · MediumCount unique detected faults without double-counting patterns.Free
  17. Q1029Floorplan a four-bank SRAM subsystemPhysical implementation · MediumFloorplan four SRAM banks around traffic, channels, pins, and power.Free
  18. Q0514First response to a macro-channel hotspotPhysical implementation · EasyDiagnose a macro-channel hotspot before applying a local fix.Free
  19. Q0103Relieve localized routing congestionPhysical implementation · EasyRelieve localized congestion without exporting timing or route risk.Free
  20. Q0081Trade speed for leakage with multi-Vt cellsPhysical implementation · EasyTrade speed and leakage with multi-Vt cells across critical paths.Free
  21. Q0391Choose a localized IR-drop repairPhysical implementation · MediumChoose a localized IR-drop repair from current and resistance evidence.Free
  22. Q0314Minimum parallel straps for IR and EM limitsPhysical implementation · HardSize parallel straps against both IR-drop and EM limits.Premium
  23. Q1040Scale distributed-wire delay with lengthPhysical implementation · EasyRelate distributed wire resistance and capacitance to length.Free
  24. Q1066Predict opposite-direction crosstalk delayPhysical implementation · EasyPredict delay when a coupled aggressor switches in the opposite direction.Free
  25. Q0702Explain the routing antenna effectPhysical implementation · EasyExplain process antenna charging and compare repair options.Premium
  26. Q0381Separate skew, jitter, latency, and uncertaintyTiming & closure · EasySeparate clock latency, skew, jitter, and timing uncertainty.Premium
  27. Q0457Compare static timing analysis with timing simulationTiming & closure · EasyCompare STA coverage with event-based timing simulation.Premium
  28. Q0475Calculate setup and hold slack with clock latencyTiming & closure · MediumCalculate setup and hold slack with explicit clock latency.Free
  29. Q0280Classify timing-path startpoints and endpointsTiming & closure · EasyClassify valid timing-path startpoints and endpoints.Free
  30. Q1095Review and scope false-path candidatesTiming & closure · MediumProve a false-path candidate is functionally impossible.Free
  31. Q0092Distinguish clock-group relationship typesTiming & closure · EasyDistinguish asynchronous, exclusive, and related clock groups.Free
  32. Q0246Choose first-order setup and hold fixesTiming & closure · EasyChoose first-order setup and hold fixes without cross-check damage.Premium
  33. Q0532Find fmax for a half-cycle timing pathTiming & closure · MediumDerive Fmax for a half-cycle launch-to-capture path.Free
  34. Q0620Use different setup and hold uncertaintyTiming & closure · MediumApply different setup and hold uncertainty budgets.Premium
  35. Q0748Explain CRPR in a timing reportTiming & closure · EasyExplain which shared clock pessimism CRPR removes.Free
  36. Q0990Explain OCV derates and temperatureTiming & closure · MediumConnect OCV derates and temperature to early and late timing.Free
  37. Q1006Pair setup and hold multicycle constraintsTiming & closure · MediumPair multicycle setup and hold constraints correctly.Free
  38. Q0879Constrain a source-synchronous read strobeTiming & closure · MediumConstrain a source-synchronous read strobe and its data window.Premium
  39. Q0272Choose coupled setup and hold ECOsTiming & closure · HardSelect coupled setup/hold ECOs that survive every required view.Free
  40. Q0929Minimum cells for two-corner closureTiming & closure · HardFind the minimum cell changes that close two opposing corners.Premium

Implementation flowMove intent through every handoff

Treat RTL-to-GDS as a chain of explicit contracts. Each stage consumes design intent, changes the implementation, and produces evidence for the next owner.

Concept model

One implementation, seven evidence gates

  1. 01RTLconstraints
  2. 02Synthesis + DFTscan-ready netlist
  3. 03Floorplanplaced macros
  4. 04Place + CTStimed clocks
  5. 05Routerouted database
  6. 06Extractparasitics
  7. 07Signofftapeout evidence
Intent in
netlist · SDC · UPF · test protocol
Evidence out
reports · checks · waivers · reproducible database
Closure loop
violation → scoped ECO → re-analysis
Logic
RTL + SDCSynthesis maps behavior while preserving clocks, exceptions, and operating assumptions.
Test
scan-ready netlistTest insertion adds controllability and observability without changing functional behavior.
Physical
placed + routedFloorplan, clocks, cells, wires, and parasitics turn logical intent into geometry.
Release
signed evidenceTiming, power integrity, physical verification, and equivalence must agree on the released database.

Synthesis, test insertion, floorplanning, placement and clocking, routing, extraction, and signoff refine the same design. A handoff is complete only when the data and its assumptions travel together.

Strong answer

A complete handoff includes the design database, the constraints and intent used to interpret it, and evidence that the producing stage met an agreed acceptance contract.

Reason it through

  • Version the netlist or layout together with timing constraints, power intent, libraries, operating conditions, and tool settings.
  • Check for missing clocks, unconstrained paths, invalid exceptions, unresolved references, and inconsistent units before downstream work begins.
  • Record quality metrics, waivers, known limitations, and an owner for every open issue so assumptions do not become invisible.
Practice a related question in the bank
Strong answer

DFT architecture starts before synthesis, scan insertion usually operates on a synthesized test-ready netlist, and its added clocks, modes, chains, and timing paths remain part of physical implementation and signoff.

Reason it through

  • Plan scan partitions, compression, memory test, test clocks, and access logic early enough to affect RTL and floorplan decisions.
  • After insertion, prove functional equivalence in functional mode and run DFT-rule checks before placement consumes the netlist.
  • Reorder chains and optimize test routing during physical design, then regenerate or validate patterns against the final implementation assumptions.
Practice a related question in the bank
Strong answer

Routing creates geometry, but tapeout requires extracted electrical analysis, manufacturing-rule checks, connectivity agreement, reliability checks, and proof that late implementation changes preserved logic.

Reason it through

  • Final parasitics can expose timing and signal-integrity failures that early estimates did not predict.
  • DRC, LVS, antenna, density, IR-drop, and electromigration checks address risks that a routed database alone does not certify.
  • A release should tie all reports, waivers, netlists, layouts, and constraints to the exact final revision.
Practice a related question in the bank

Structural testControl, capture, and observe faults

Scan converts difficult sequential state into controllable and observable test state. ATPG still needs a sensitized path and a capture event that preserves the fault effect.

Concept model

Separate shift time from capture time

S0Shiftscan_en=1Load state
S1Shiftscan_en=1Load state
LLaunchscan_en=0Launch transition
CCapturescan_en=0Sample response
U0Unloadscan_en=1Observe state
U1Unloadscan_en=1Observe state

Illustrative launch-on-capture sequence; launch-on-shift changes the launch edge, not the activation–propagation–observation obligation.

ModelActivateObserveTargets
Stuck-atforce 0 or 1single capturelogical defect
Transitionlaunch edgeat-speed captureslow-to-rise/fall
Shift
SE = 1Serial clocks load stimulus and unload a prior response through each scan chain.
Capture
SE = 0Functional logic launches or samples the response on the defined test-clock edge.
Compact
many → fewResponse compactors reduce output volume but require deliberate handling of unknown values.
Diagnose
pattern + chainFailing cycles and chain positions help localize likely defect sites after detection.

Shift cycles load and unload scan state with scan enable asserted. One or more functional-speed capture edges launch and sample the response before the chain returns to shift mode.

Strong answer

Shift mode connects scan flops as serial chains to load and unload state. Capture mode restores the functional data paths so one or more controlled clock edges record the circuit response.

Reason it through

  • Scan enable and any compression controls must settle before the clock protocol changes modes.
  • Stuck-at patterns commonly use one capture edge, while transition patterns use launch and capture events at the intended functional speed.
  • The tester compares unloaded response bits with expected values after accounting for masked or intentionally unknown positions.
Practice a related question in the bank
Strong answer

ATPG drives the fault site to the opposite of the faulty value, chooses side inputs that sensitize a path, and propagates the resulting difference to a scan cell or observable output.

Reason it through

  • Activation, propagation, and observation are separate requirements; satisfying only one does not detect the fault.
  • Controlling values on reconvergent logic, sequential depth, X sources, and test constraints can block propagation.
  • Untestable and aborted faults should be classified by reason so a coverage gap can be accepted or repaired deliberately.
Practice a related question in the bank
Strong answer

A transition test launches a value change and captures its arrival using the functional-speed edge spacing, so excessive path delay can be observed as an incorrect sampled value.

Reason it through

  • Launch-on-capture and launch-on-shift protocols create different initialization and clock-control obligations.
  • The targeted path needs a valid transition, sensitized side inputs, and a capture point that remains visible through compaction.
  • Test clocks, clock gating, on-chip clock controllers, and tester timing must reproduce the intended launch-to-capture interval.
Practice a related question in the bank

Embedded testGive memories and test logic clear ownership

Memory BIST, scan compression, and JTAG solve different access problems. Their controllers, clocks, resets, and result paths must compose without contention.

Concept model

Route one test request to the right engine

External accessJTAG TAPinstruction + data registers
Logic testscan controllerchains · compression · ATPG
Memory testMBIST controlleraddress · data · compare
responseX-maskMISRexpected signature
  1. w0
  2. r0 · w1
  3. r1 · w0
  4. r0 · w1
  5. r1 · w0
  6. r0
Logic test
scan + ATPGChains expose sequential state so structural patterns can target random logic faults.
Memory test
March elementsOrdered reads and writes expose cell, transition, address, and coupling behaviors.
Compression
MISRA deterministic recurrence condenses response streams into a final signature.
Access
TAP / IJTAGInstructions select data registers or embedded instruments without giving two engines control at once.

A TAP or chip test controller selects an instrument. Scan logic tests random logic, MBIST applies memory-specific operations, and status returns through an owned, deterministic path.

Strong answer

An unknown entering a linear compactor can contaminate many signature bits and hide otherwise deterministic fault effects, so known X sources need masking, blocking, control, or bypass before compaction.

Reason it through

  • Common sources include uninitialized memories, analog boundaries, uncontrolled clocks, and nondeterministic functional state.
  • Masking protects observability but can reduce coverage, so ATPG must know which channels and cycles are unavailable.
  • X-bounding logic and masks need their own structural checks because a stuck mask can silently discard valid response data.
Practice a related question in the bank
Strong answer

A March test applies ordered read and write operations while traversing addresses in specified directions, creating the background and neighbor activity needed to expose cell, transition, address, and coupling faults.

Reason it through

  • Each read checks the value established by a previous element before the next write changes the background.
  • Ascending and descending passes exercise address order and opposite neighbor relationships.
  • The controller must define operation order, read latency, repair behavior, and the exact response used to set fail status.
Practice a related question in the bank
Strong answer

Scan and ATPG target random logic, MBIST applies memory-specific algorithms, and JTAG or an embedded-instrument network provides external access and selection. A chip test controller coordinates shared clocks, resets, modes, and status.

Reason it through

  • Every test instruction needs a defined data path, active controller, legal clock source, and exit sequence back to functional operation.
  • MBIST may own a memory interface during test, so isolation and muxing must prevent concurrent functional access.
  • Bypass, reset, and unsupported-instruction behavior should be deterministic so one instrument cannot strand another.
Practice a related question in the bank

Physical foundationMake placement and clocking physically possible

A useful floorplan reserves routing, power, and clock resources before placement optimizes local timing. Macro geometry and pin access often dominate the first-order result.

Concept model

Read the floorplan as a resource map

clock sourceroot buffer
branch Abranch B
FF0FF1FF2FF3

Balance insertion delay while controlling skew, slew, power, and routing demand.

Macros
place by trafficKeep strongly connected blocks close while preserving channel width and legal pin access.
Density
local mattersA safe block average does not prevent a hotspot near macro edges or narrow corridors.
Clock
latency + skewCTS trades insertion delay, sink balance, transition, power, and route resources.
Evidence
early routeGlobal routing and trial CTS expose structural problems before detailed implementation.

Macro halos and channels protect routing access, edge pin placement controls long crossings, the power grid consumes tracks, and the clock tree must reach every sink with bounded latency and skew.

Strong answer

A routable floorplan gives high-connectivity regions enough tracks, pin access, vias, and detour options after accounting for macros, halos, blockages, the power grid, and clock resources.

Reason it through

  • Place macros from connectivity and interface direction, then reserve channels based on pin demand and available routing layers.
  • Avoid narrow notches, aligned macro pin walls, and macro corners where local demand can exceed capacity.
  • Validate with early global routing and pin-access analysis because total core utilization cannot reveal local shortages.
Practice a related question in the bank
Strong answer

Utilization measures placed-cell area over a region, while congestion compares routing demand with local track and via capacity. Macros, pins, blockages, and directional traffic can exhaust capacity even when cell area is modest.

Reason it through

  • Macro channels and high-pin-density edges can create localized demand that the block average hides.
  • Power straps, clock routes, reserved layers, and routing blockages reduce the capacity available to signal nets.
  • Useful fixes include spreading cells, moving macros or pins, adjusting blockages, improving hierarchy, and changing layer allocation.
Practice a related question in the bank
Strong answer

CTS builds a clock network that reaches every sink within transition, capacitance, latency, skew, routing, and power limits across required scenarios. The best tree is the one that supports full-path timing, not necessarily the one with minimum skew.

Reason it through

  • Balancing sink arrival can improve one path group while insertion delay and added buffers increase power and uncertainty elsewhere.
  • Useful skew is intentional only when setup and hold consequences are checked across all interacting paths and corners.
  • Clock gates, generated clocks, nondefault routing, shielding, and local density all affect achievable topology.
Practice a related question in the bank

Interconnect integrityClose wires, coupling, and current together

Post-route behavior depends on distributed resistance, capacitance, coupling, and the voltage delivered by the power grid. A timing-only wire fix can move congestion or power risk elsewhere.

Concept model

Two coupled nets share a power delivery problem

aggressorswitches ↑
Ccouplesame direction can speed; opposite direction can slow
victimΔdelay
IR drop
ΔV = I · R
EM
J = I / area
Repair
wider · shorter · more vias/straps
Wire RC
length hurtsResistance and capacitance grow with geometry, so long unbuffered routes can dominate path delay.
Coupling
direction mattersOpposite switching can increase effective load, while same-direction switching can reduce it.
IR drop
V = I · RGrid resistance and simultaneous current lower the voltage available at cell rails.
EM
current densityWire and via geometry must carry required average and peak current within reliability limits.

Long signal routes accumulate RC delay, neighboring transitions change effective coupling, and switching current creates voltage loss in the grid. Signoff uses extracted geometry to analyze all three.

Strong answer

A long route accumulates distributed resistance and capacitance, while coupling capacitance changes the effective load according to relative aggressor and victim switching. Opposite-direction transitions usually slow the victim more than quiet or same-direction neighbors.

Reason it through

  • Buffering divides a long RC line into shorter stages but adds cell delay, power, placement demand, and new timing arcs.
  • Spacing, layer changes, shielding, and slew control reduce coupling when route topology is the root cause.
  • Extraction-aware SI analysis determines whether aggressors overlap in a valid timing window rather than assuming worst-case switching everywhere.
Practice a related question in the bank
Strong answer

IR drop limits the voltage lost through resistive paths at required current, while electromigration limits current density in wires and vias. Strap and via sizing must satisfy both constraints in every required activity scenario.

Reason it through

  • A wide strap can reduce resistance, but current still needs enough vias and low-resistance paths through every grid transition.
  • Static analysis emphasizes average demand; dynamic analysis captures localized simultaneous switching and transient droop.
  • Repairs include stronger straps, denser meshes, more vias, closer power sources, cell spreading, decap, and lower local switching demand.
Practice a related question in the bank
Strong answer

During fabrication, charge collected on a partially formed conductor can stress a connected gate oxide before a safe discharge path exists. Antenna rules bound exposed conductor area relative to vulnerable gate area for each process stage.

Reason it through

  • A layer jump can break the risky conductor segment by moving part of the route to a layer formed later in the process.
  • An antenna diode provides a discharge path, but it adds junction capacitance, leakage, placement demand, and routing.
  • Repairs must be rechecked against timing, congestion, DRC, and the foundry rule that identified the violation.
Practice a related question in the bank

Timing contractsConstrain the sampled path, not the picture

Static timing compares data arrival against a required sampling window for every declared scenario. Clocks, exceptions, uncertainty, and variation define that window.

Concept model

Join launch, data, and capture on one ruler

  1. 0.0 nsLaunch edgeclock + tCQ
  2. 0.3 nsData startscell + wire delay
  3. 2.5 nsData arrivesarrival time
  4. 3.0 nsCapture edgerequired time
Setupslack = required − arrivalmax path · next capture edge
Holdslack = arrival − requiredmin path · same capture edge
Skew
capture − launch latency
Uncertainty
jitter + modeling margin
OCV
early/late path derates
CRPR
remove common-path pessimism
Setup
late dataMaximum-delay analysis asks whether data settles before the intended capture edge.
Hold
early dataMinimum-delay analysis asks whether old data remains stable after the active edge.
Skew
capture − launchPositive capture skew usually helps setup and hurts hold for the same edge relationship.
CRPR
shared clock pathCommon-path pessimism removal avoids charging incompatible variation twice on shared clock segments.

Setup protects the next capture boundary and hold protects the current one. Clock latency, skew, uncertainty, derates, and common-path pessimism change the two checks in different ways.

Strong answer

Setup checks a maximum-delay path against a later capture boundary, while hold checks a minimum-delay path against the current boundary. Speeding data helps setup but can hurt hold; delaying data helps hold but can consume setup margin.

Reason it through

  • Setup fixes include reducing logic depth, upsizing critical cells, buffering fanout, improving placement, or using safe clock skew.
  • Hold fixes usually add localized minimum delay or adjust clock arrival without changing the functional cycle relationship.
  • Every ECO needs multi-corner analysis because the setup-critical and hold-critical corners and paths can differ.
Practice a related question in the bank
Strong answer

Skew changes the relative launch and capture arrivals, uncertainty reserves margin for unmodeled clock variation, OCV derates model different path behavior, and CRPR removes impossible double counting on shared clock segments.

Reason it through

  • Positive capture skew generally adds setup time and removes hold margin for the same launch-capture relationship.
  • Setup and hold can use different uncertainty and derate values because their edge relationships and variation risks differ.
  • CRPR removes only pessimism associated with a physically common clock path; it does not erase real clock variation.
Practice a related question in the bank
Strong answer

An exception is valid only when it describes a real architectural sampling relationship: a false path is never functionally captured, a multicycle path is intentionally captured on a later edge, and clock groups declare defined clock relationships or their absence.

Reason it through

  • A setup multicycle usually needs a paired hold adjustment so the minimum-delay obligation remains on the intended edge.
  • Asynchronous clock groups remove cross-domain timing analysis only when a separate CDC structure and verification plan make that safe.
  • Exceptions should be narrow, reviewable, and matched to real startpoints, endpoints, and modes rather than broad naming patterns.
Practice a related question in the bank

Release evidenceClose every required scenario on one database

Tapeout readiness is an intersection, not a best result. Timing, signal integrity, power integrity, physical verification, and logical equivalence must all pass their required scope.

Concept model

Turn violations into bounded ECO loops

ModeCornerDominant checkState
FunctionalSS · low V · hotSetupreview
FunctionalFF · high V · coldHoldpass
Scan shiftSS · low V · hotIR / slewrepair
At-speed testFF · high V · coldCapture holdpass

Illustrative obligations only: dominant PVT and RC views depend on the libraries, test protocol, extracted design, and signoff methodology.

Setupsize · buffer · route · pipeline
Holddelay cell · route detour
SIspace · shield · resize
IR / EMstraps · vias · spread load
DRC / antennareroute · diode · layer hop
LVS / equivalenceconnectivity or netlist ECO
MMMC
all required viewsFunctional and test modes combine with voltage, process, temperature, and RC corners.
Physical
DRC + LVSGeometry must satisfy manufacturing rules and match the intended circuit connectivity.
Electrical
STA + SI + PITiming, coupling, voltage delivery, and current density use extracted implementation data.
Logical
equivalenceThe released implementation must remain functionally consistent with its reference design.

Classify the failing scenario, choose a root-cause fix, rerun the affected analyses, and then restore full signoff coverage. The released netlist and layout must remain logically and physically consistent.

Strong answer

MMMC closure means every required analysis view meets its acceptance criteria on the same releasable design, with modes, clocks, parasitics, libraries, voltage, process, temperature, derates, and exceptions configured consistently.

Reason it through

  • Functional, scan shift, at-speed test, and low-power modes can activate different clocks, paths, and constraints.
  • Maximum and minimum timing, SI, and power-integrity risks can peak in different process, voltage, temperature, and RC combinations.
  • Scenario reduction is safe only when dominance or equivalence is demonstrated and the reduced set remains documented.
Practice a related question in the bank
Strong answer

First confirm the violation and its model, then choose a localized root-cause fix, run fast incremental analysis, and finally reestablish full timing, physical, power, and equivalence signoff on the updated database.

Reason it through

  • Check clocks, exceptions, parasitics, noise, and scenario setup before modifying cells or routes.
  • Predict setup, hold, slew, capacitance, congestion, leakage, dynamic power, and cross-corner effects before selecting an ECO.
  • After implementation, rerun extraction and all affected analyses, then perform clean full-suite signoff rather than relying only on incremental reports.
Practice a related question in the bank
Strong answer

STA checks declared timing relationships, SI checks coupling effects, IR and EM check power delivery and conductor reliability, DRC checks geometry, LVS checks connectivity, and equivalence checks logical consistency with the reference.

Reason it through

  • The analyses are complementary: clean LVS does not prove timing, and clean STA does not prove manufacturable geometry or reliable power delivery.
  • A waiver needs a bounded rule, location, rationale, owner, and approval; it is not the same as a passing result.
  • All reports must reference the same final netlist, layout, extraction, constraints, libraries, and tool-qualified runset.
Practice a related question in the bank

Continue preparing