Qualcomm interviews for chip design roles probe whether you understand the flow you work in, not just the tool commands. For physical design, expect questions on timing constraints, clock tree synthesis, scan chains, multi-Vt cells, power, IR drop and signal integrity, usually followed by "why" and "what would you do if" follow-ups. This post collects questions reported from Qualcomm interviews, grouped by area, with model answers you can check your own against.
Key takeaways
- Every I/O port needs a timing constraint, or its paths are simply not checked.
- Know the difference between ideal clocks before CTS and propagated clocks after it, and what that does to latency and uncertainty.
- Lock-up latches, clock gating cells and multi-Vt swaps are favorite topics because they connect timing, test and power.
- Be ready to write real SDC syntax and to estimate wire RC on paper.
- Answer with trade-offs and numbers, not definitions alone.
The topics in plain words
- Timing constraints (SDC): a file that tells the tools how fast each clock runs and when signals arrive at and leave the block. Anything left unconstrained is simply not checked.
- Clock tree synthesis (CTS): building the network of buffers that delivers the clock to every flip-flop at nearly the same time. Before it exists, tools assume an ideal clock; after it, they measure the real one.
- Scan chains: for manufacturing test, flip-flops are linked into long shift registers. Placing and reordering them, and keeping them safe for hold timing, is a physical design job.
- Multi-Vt cells: the same logic cell in fast-but-leaky and slow-but-frugal versions. Choosing between them trades speed against standby power.
- Power and IR drop: clock gating and power gating save energy; a strong power grid and decoupling capacitors stop the supply voltage sagging where many cells switch at once.
- Signal integrity: neighboring wires couple, so a switching wire can slow, speed up or put a glitch on its neighbor. Noise and glitch reports list where that is a risk.
RTL Design
Work in progress. Questions for this section are still being collected. If you have been asked RTL design questions in a Qualcomm interview, please share them in the comments so they can be added.
Design Verification
Work in progress. Questions for this section are still being collected. If you have been asked verification questions in a Qualcomm interview, please share them in the comments so they can be added.
How interviewers probe these answers
- After "constrain all I/O", expect: what do you do with a port that has no clear clock?
- After lock-up latches, expect: draw the waveform and show where the half cycle comes from.
- After multi-Vt, expect: how do you decide which cells to swap, and what happens to hold when you swap to LVT?
- After pre- and post-CTS, expect: why can hold get worse after CTS even when skew is low?
Frequently asked questions
Are these the exact questions Qualcomm asks?
They are questions reported from interviews. Each team sets its own questions, so treat them as a guide to topics and depth.
Which tools should I know?
Know the concepts first. Interviewers expect familiarity with at least one implementation and one sign-off timing tool, but they care more about why a command is used than its exact options.
How should I prepare for the physical design round?
Be able to explain the full flow from netlist to sign-off, write basic SDC by hand, and talk through one real timing or congestion problem you solved, with numbers.
How to structure an answer
A strong answer to any question below has three parts: a one-sentence definition, the reason it matters for timing, power or yield, and what you would actually do or check in the tool. For example, for clock gating: it stops the clock to idle registers to save dynamic power; it must be glitch-free, which is why an integrated clock gating cell with a latch is used; and in CTS you check the enable timing at the gate, which is tighter because the gate sits earlier in the tree. Interviewers then push on the third part with a "what if", so add a number wherever you can.
Deep dive: the numbers and details
This part holds the 24 physical design questions with model answers, SDC examples, an RC estimate and a power calculation. Skip it if the topic overview is all you need.
1. In building the timing constraints, do you need to constrain all I/O (input-output) ports?
Yes. A port with no input or output delay has no timing reference, so the paths from or to it are unconstrained and static timing analysis will not check them. Every functional input needs set_input_delay and every output needs set_output_delay against the clock that launches or captures it outside the block, along with a driving cell on inputs and a load on outputs so that transition and delay are realistic. Ports that are truly asynchronous or static, such as a reset synchronizer input or a mode pin, should be handled explicitly with a false path or a case analysis value, never just left out. Complete I/O constraints are what make the interface to the rest of the chip, or to the board, reliable: each signal arrives in time at the flip-flop that captures it. Most tools can list unconstrained ports, for example with check_timing, and that list should be empty or explained before sign-off.
A note for FPGA engineers: in FPGA flows, I/O constraints also include pin assignments, which place a signal on a specific pin or I/O bank and set its I/O standard. Those are placement constraints, separate from the timing constraints discussed here.
2. Can a single port be multi-clocked? How do you set delays for such ports?
Yes, a single port can be associated with more than one clock. For example, a port can work at two frequencies, a regular clock and the same clock divided by two, or it can be shared between two interfaces in different modes. Add one delay per clock and use -add_delay so the second command does not overwrite the first:
set_input_delay -clock clk_fast 1.2 [get_ports din]
set_input_delay -clock clk_div2 2.5 -add_delay [get_ports din]
If the two clocks are never active on the port at the same time, define them as exclusive with set_clock_groups -logically_exclusive or analyze each mode in its own scenario, so the tool does not time impossible cross-clock paths.
3. How is scan DEF (Design Exchange Format) generated?
Scan DEF is written by the DFT insertion tool after scan chains are stitched, for example with write_scan_def in Synopsys DFT Compiler or write_scandef in Cadence Genus, and read into the place and route tool alongside the netlist. The file holds the scan chain connectivity information, which is also present in the netlist. Some EDA tools can auto-trace the scan chains without a scan DEF, given the scan-in and scan-out points. However, scan DEF carries extra information that makes it easier to map scan chains. For example, when there are multi-input components in the chain, tracing becomes difficult. Auto-tracing also cannot recover PARTITION information (scan flops can be swapped across chains in the same PARTITION). The placer uses this to reorder chains by physical location, which cuts scan wirelength and congestion. These advantages make scan DEF the preferred method over auto-tracing.
4. What is the purpose of a lock-up latch in a scan chain?
A lock-up latch is a transparent latch placed between two scan flops where clock skew is large and meeting hold timing during scan shift is a challenge because of long uncommon clock paths. A typical case is a chain that crosses from one clock domain to another. Functionally the two domains may never interact, so their clocks are not balanced against each other and have a large uncommon path. In scan shift mode, however, data moves from one domain to the next, and a hold failure becomes likely. Without lock-up latches it would be very hard for the STA engineer to close timing on scan chains across domains, and the large uncommon path would also expose the chain to large on-chip variation.
For positive-edge flops, the latch is clocked by the launching flop's clock and is transparent while that clock is low. It holds the launched data for half a cycle, which adds roughly half a clock period of hold margin at the receiving flop at the cost of a little setup margin on a slow shift clock. That is why lock-up latches are often called the soul mate of scan-based designs.
5. Explain short circuit current.
In CMOS, short circuit current is the momentary current that flows from supply to ground when the nMOS and pMOS networks are both partly on during a transition, while the input is between the nMOS threshold and VDD minus the pMOS threshold. It causes a brief blip of current. When input and output rise and fall times are comparable, it is generally less than 10% of dynamic power. It grows with slow input transitions, which is one reason max transition limits matter for power as well as timing.
6. What are the pros and cons of using different Vt cells?
Using cells with different threshold voltages (Vt) lets designers trade speed against leakage. Threshold voltage is set by process choices such as channel doping or, in FinFET and newer processes, work function metal. Common flavours are:
| Flavour | Speed | Leakage | Typical use |
|---|---|---|---|
| ULVT (ultra-low Vt) | Fastest | Highest | Only the most critical paths |
| LVT (low Vt) | Fast | High | Timing-critical paths |
| SVT or RVT (standard Vt) | Medium | Medium | Default balance of power and timing |
| HVT (high Vt) | Slowest | Lowest | Non-critical logic, always-on domains |
Ultra-low Vt cells turn on at a lower gate voltage, which gives faster switching. Their cost is much higher leakage, which raises static power, so designers use them sparingly and balance them against the other flavours. They do not reduce power by themselves; the only power gain is indirect, when the extra speed lets the design run at a lower supply voltage.
In timing-critical paths, LVT cells can be used to take advantage of their higher performance. For the clock tree, use a single Vt flavour to keep variation between branches low. The choice depends on the design's timing and power targets; many teams reserve SVT cells for clock trees, while some high-performance designs use LVT. Different Vt versions of the same cell usually share one footprint, so they can be swapped late in the flow without disturbing placement. The timing impact of each swap still has to be checked, and implementation tools normally do that analysis automatically.
7. How do you set inter-clock uncertainty?
Inter-clock uncertainty is the margin applied to paths that launch on one clock and capture on another. It covers skew and jitter between the two clocks. It is set with set_clock_uncertainty using -from and -to, and can differ for setup and hold:
set_clock_uncertainty -setup 0.5 -from [get_clocks clock1] -to [get_clocks clock2]
set_clock_uncertainty -hold 0.1 -from [get_clocks clock1] -to [get_clocks clock2]
The value is in the time unit of the loaded library, usually nanoseconds, so 0.5 means 500 ps. SDC has no option to give uncertainty as a percentage of the clock period; if a team works that way, the script must compute the value.
8. In Design Compiler, how do you constrain clocks, I/O ports, max capacitance and max transition?
In Synopsys Design Compiler (DC), as in any SDC-based tool:
# 10 ns clock on port clk
create_clock -name clk -period 10 [get_ports clk]
# 2 ns input delay and 1 ns output delay relative to clk
set_input_delay -clock clk 2 [get_ports data_in]
set_output_delay -clock clk 1 [get_ports data_out]
# maximum capacitance on an output port, in library units (often pF)
set_max_capacitance 0.05 [get_ports data_out]
# maximum transition on a port, or on the whole design
set_max_transition 1 [get_ports data_out]
set_max_transition 0.5 [current_design]
Design-wide limits applied to current_design act on every net; tighter limits on individual ports or clocks override them. The library also carries its own default limits, and the tighter of the two applies.
9. What are the differences in clock constraints from pre-CTS to post-CTS?
Clock tree synthesis (CTS) inserts buffers and inverters to distribute the clock and balance its skew. Before CTS, there is no real tree, so clocks are treated as ideal, and the constraints stand in for the tree that does not yet exist: set_clock_latency gives an estimated insertion delay, set_clock_uncertainty includes an allowance for skew on top of jitter and margin, and set_clock_transition sets a realistic slew at the clock pins. After CTS, the tree exists, so the clocks are switched to propagated with set_propagated_clock: the tool calculates actual latency, skew and transition through the tree. The estimated network latency is removed (source latency outside the block stays), the ideal transition is removed, and uncertainty is reduced to jitter and sign-off margin because skew is now measured. Post-CTS analysis then confirms that the design still meets setup and hold with the real tree.
10. How is clock gating done?
Clock gating cuts dynamic power by stopping the clock to registers whose value is not changing. In practice synthesis tools infer it from enable logic: a register bank with a load enable, which would otherwise use a feedback multiplexer, is clocked through an integrated clock gating (ICG) cell instead. An ICG cell combines a latch with an AND gate (for positive-edge logic). The latch is transparent while the clock is low, so the enable can only change while the clock is low, and the gated clock cannot glitch. A plain AND gate without the latch is unsafe, because an enable changing while the clock is high would produce a clipped clock pulse. Clock gating can be applied at any granularity, from a few flip-flops to whole blocks; coarse gating saves more clock tree power, fine gating catches more idle registers. Tools usually set a minimum bank size so a gate is not inserted for one or two flops.
11. What constraints do you add in CTS for clock gates?
The key issue is the enable timing check at the ICG. The ICG sits earlier in the clock tree than the flops it drives, so its clock arrives earlier and the enable path has less time. Typical measures: place ICGs close to their sinks or let the tool clone them; tell CTS to balance through ICGs to their sinks rather than stop at them; set clock gating setup and hold checks (these are inferred automatically for library ICG cells); limit fanout per ICG; and keep ICGs in the same Vt flavour and drive strength range as the rest of the tree.
12. What is the trade-off between dynamic power and leakage power?
Dynamic power is roughly activity times capacitance times VDD squared times frequency, so it falls quickly with lower supply voltage. Leakage rises sharply as threshold voltage drops, roughly a decade for every 70 to 100 mV of Vt reduction at room temperature. Lowering VDD saves dynamic power but slows the logic; recovering the speed with lower-Vt cells costs leakage. The best point depends on activity: busy logic is dominated by dynamic power, while idle or rarely used logic is dominated by leakage.
13. How do you reduce standby (leakage) power?
Power gating with header or footer switches, plus isolation cells on outputs and retention flops for state that must survive; HVT or longer-channel cells on non-critical paths; reduced retention voltage for memories; reverse body bias where the process supports it; and turning off unused analog blocks and PLLs.
14. Explain the top-level pin placement flow. What parameters decide it?
Pins follow the package or bump plan and the positions of neighboring blocks. Key parameters: which side and layer each bus should leave from, keeping bus bits together and in order, timing-critical pins close to their logic, pin pitch and spacing rules, keeping pins away from macros and corners, shielding or spacing for sensitive and high-speed signals, and power and ground bump or pad positions.
15. Given block netlists, timing constraints, libraries and macro LEFs, how will you start floorplanning?
First check the inputs: the netlist links with no missing cells, the constraints load cleanly, and check_timing is clean. Estimate the core area from cell and macro area at a target utilization, often 60 to 75% to start. Place macros along the edges following the data flow, with pins facing the core, channels between them and halos around them. Place I/O pins, build the power grid and add placement blockages. Then run a quick trial placement and look at congestion and timing before committing.
16. With a net length of 1000 um, how will you compute RC values?
Use per-unit-length values from the technology file or the extraction rule deck for the chosen layer and width. Resistance is sheet resistance times the number of squares (length divided by width). Capacitance is the per-micron capacitance, including coupling, times length.
Illustrative example. A wire 0.1 um wide with 0.1 ohm per square sheet resistance has 1000 / 0.1 = 10,000 squares, so R = 1,000 ohm. At 0.2 fF/um, C = 200 fF. RC = 1,000 x 200e-15 = 200 ps. The 50% delay of a distributed RC line is about 0.38 RC, so about 76 ps of wire delay, before driver and load effects. That is long enough that a repeater would normally be added. These numbers are illustrative, not from a specific process.
17. What do noise reports represent?
They report crosstalk effects. Glitch noise is a bump coupled onto a quiet victim net when an aggressor switches; delta delay is the change in delay of a switching victim when aggressors switch with or against it. Noise reports list victim nets whose glitch could exceed the noise immunity of the receiving cell, and timing reports with signal integrity enabled include the delta delays.
18. What do glitch reports contain?
For each victim net: the aggressor nets, the glitch peak height and width, the receiver's noise immunity limit, and the slack between the two. Glitches on clock, reset and latch enable nets get the most attention, because they can change state directly.
19. What are the CTS steps in IC Compiler?
In the original Synopsys IC Compiler flow: define clock tree options (skew and latency targets, buffer and inverter lists, max transition, max capacitance, non-default routing rules such as double width and spacing), check the clock tree with check_clock_tree, build it with clock_opt -only_cts, optimize with clock_opt -only_psyn, route the clock nets, and report with report_clock_tree and report_clock_timing. IC Compiler II and Fusion Compiler follow the same ideas through clock_opt stages.
20. What does a clock constraints file contain?
Clock definitions (create_clock, create_generated_clock), latency, uncertainty, transition, clock groups for asynchronous or exclusive clocks, and the propagated clock setting. A CTS specification adds skew targets, buffer lists, routing rules and exception pins, such as pins where the tree should stop or be ignored.
21. How do you analyze clock tree reports?
Compare against the targets: maximum and minimum latency, global and local skew, transition and capacitance violations, number of levels, buffer count and clock power. Then look for outliers, such as one branch far longer than the rest, sinks that were excluded by mistake, or big latency differences between sinks that talk to each other.
22. What do VoltageStorm IR drop reports represent?
VoltageStorm was Cadence's power grid analysis tool. Its engines moved into the Encounter Power System at the end of 2008, as a Cadence blog post explains, and Cadence introduced the Voltus IC Power Integrity Solution in November 2013, its current power integrity product. Its reports show the voltage drop from the supply pins to each instance, as a map and as a list of worst instances, for static (average current) and dynamic (switching) analysis, along with electromigration violations on power rails. Each instance's drop is compared with the budget the design allows, and hot spots lead to grid fixes, cell spreading or decap insertion.
23. Where and when do you use DCAP (decoupling capacitor) cells?
Near cells with large simultaneous switching current, such as clock buffers, high-drive cells, and logic next to macros and I/O, to supply local charge and reduce dynamic IR drop. They are added after placement and before or with filler cells. They cost area and some gate leakage, so they go where the dynamic IR analysis shows they are needed rather than everywhere.
24. What are various power reduction techniques?
Clock gating, multi-Vt cell use, power gating, multiple voltage domains and dynamic voltage and frequency scaling (DVFS), operand isolation, memory low-power modes, lower switching activity through architecture and encoding, and right-sized cells and shorter wires to reduce capacitance.
25. Worked example: what a supply voltage cut buys
Interviewers often follow question 12 with numbers. Dynamic power is P = α · C · VDD2 · f. Take an activity factor α = 0.1, a total switched capacitance C = 1 nF, VDD = 0.8 V and f = 1 GHz: P = 0.1 × 1 nF × 0.64 V2 × 1 GHz = 64 mW. Dropping VDD to 0.7 V gives 0.1 × 1 nF × 0.49 × 1 GHz = 49 mW, about 23% less, for a 12.5% voltage cut. The catch is speed: the logic is slower at 0.7 V, and if lower-Vt cells are swapped in to recover it, leakage rises, which is the trade-off in question 12. These values are illustrative, not from a specific product.
Further reading
- Cadence blog, VoltageStorm is alive and kicking: how VoltageStorm's engines moved into the Encounter Power System launched at the end of 2008.
- EEJournal, Cadence Introduces Voltus IC Power Integrity Solution (November 12, 2013), and the current Voltus IC Power Integrity Solution product page.
- David Harris, CMOS VLSI Design, lecture 10: Sequential Circuits (Harvey Mudd College, 2004): setup, hold and skew equations, and latch timing, which underpin the lock-up latch answer.


Your comments will be moderated before it appears here.