← RTL Design Interviews

Interview Question · Design

Doubling Throughput: A Micro-Architecture Call

L3 SeniorL4 Staffmicro architecturetradeoffspipelining

The Question

The interviewer asks

Your packet-processing block meets timing at 1 GHz but must now handle twice the traffic. Walk through your options and how you would choose.

Reveal the Model Answer

Start with questions: is the limit the clock, the datapath width, or dependencies between packets? Then compare three options. A deeper pipeline adds latency and hazards. Two lanes double area and need ordering at the merge. A wider datapath processes two units per cycle, but needs alignment logic. Choose by area, power, latency budget and verification effort, and prototype the risky part first.

Three Ways to Double ThroughputArrows carry packets. Each middle box is one option; a design picks one of them.
Three ways to double throughputArrows carry packets. Each middle box is one option; a design picks one of them. component Packets in the stream toprocess component Deeper pipeline more stages: latencyand hazards component Two lanes two copies: doublearea component Wider datapath two units per cycle:alignment logic component Merge in order tags restore packetorder component Packets out same order as theycame in split
  1. Deeper

    A shorter stage allows a faster clock. Each new stage adds a cycle of latency and a hazard to check.

  2. Lanes

    Two copies work on alternate packets. The merge must put the results back in order.

  3. Wider

    One copy handles two units per cycle. Packets that do not fill a word need alignment logic.

Read the diagram as text
  • Packets in: the stream to process
  • Deeper pipeline: more stages: latency and hazards
  • Two lanes: two copies: double area
  • Wider datapath: two units per cycle: alignment logic
  • Merge in order: tags restore packet order
  • Packets out: same order as they came in
  • Packets in → Deeper pipeline
  • Packets in → Two lanes: split
  • Packets in → Wider datapath
  • Two lanes → Merge in order
  • Deeper pipeline → Packets out
  • Merge in order → Packets out
  • Wider datapath → Packets out

Each option costs something different: latency, area or alignment logic. The table below compares them.

OptionGainsCostsMain risk
Deeper pipeline, faster clockSmall areaLatency, more flops, clock powerHazards and stalls between dependent packets
Two parallel lanesKeeps the clock and per-packet logicAbout twice the areaReordering at the merge
Twice the datapath widthSame clock, one control pathWider muxes and alignmentPackets that end mid-word

A senior answer chooses with numbers: area and power budgets, the latency limit, and what the verification plan must add.

Self-Check

Tick each key point you covered in your answer.

What Interviewers Listen For (Senior)

A good senior answer asks what limits throughput today before choosing. It names one cost and one verification risk for each option.

Cross-Questions

Where the interviewer goes next. Open a node to practise that follow-up.

What IfProve ItCompareScale It

  1. Doubling Throughput: A Micro-Architecture Call
  2. CompareA Wider Datapath Versus Two LanesL3 Senior

    Compare doubling the datapath width with building two independent lanes.

    Reveal the Model Answer

    A wider datapath keeps one control path and one ordering, but packets rarely fill whole words. So it needs alignment and partial-word handling. Two lanes keep the narrow logic, but each lane needs control and the merge must restore order. Short packets favour lanes; long streaming data favours width.

    Self-Check

    Tick each key point you covered in your answer.

    Missed Corner

    "Width is free if the bus is already wider."

    The processing logic must handle two units at once, including a packet ending in the first half. That rare case is where bugs hide.

    Source Lesson R8.3 review claims (AXI-Stream filter: TKEEP and packet ends)

  3. Scale ItThe Verification Cost of an Architecture ChoiceL3 SeniorL4 Staff

    How does each option change the verification plan, and how should that affect the choice?

    Reveal the Model Answer

    A deeper pipeline adds hazard and stall scenarios, which need directed tests and coverage of back-to-back dependencies. Two lanes add ordering checks at the merge and tests that make the lanes finish out of order. A wider datapath adds alignment corners. Count that work in weeks and put it next to the area and power numbers.

    Self-Check

    Tick each key point you covered in your answer.

    Wrong Altitude

    "Verification will adapt to whatever we build."

    The verification effort is often the schedule. An option that saves 5% area but adds a reorder buffer can cost more time than it saves.

  4. What IfWhen Packets Depend on Each OtherL3 Senior

    Asked if they choose the deeper pipeline

    What if each packet updates a counter table that the next packet may read?

    Reveal the Model Answer

    Then a deeper pipeline creates a read-after-write hazard. The next packet may read the table before the earlier update has been written back. You need forwarding of the in-flight update, or a stall when the same entry is used. Both add logic and rare cases that the verification plan must cover.

    Self-Check

    Tick each key point you covered in your answer.

    Half-Truth

    "Pipelining never changes the function."

    It does not change the function of a pure datapath. With shared state, a read can overtake a write that is still in the pipe.

    Source Lesson R4.6 review claims (registers cannot be moved into a feedback loop without changing the result)

  5. Prove ItSolve R5.6.1: A Merger That Favours A Two to OneSimulator-graded

    Show it, don't just say it: this Silicon Lab problem is graded by the simulator.

    Open Problem R5.6.1

What People Get Wrong

Common wrong answers and why they fail.

Wrong Altitude

Answers "pipeline it deeper" at once.

A deeper pipeline only helps if the clock can rise and packets are independent. Without asking, the candidate may solve the wrong limit.

Delivery

Lists the three options with no way to choose between them.

The interviewer wants the decision rule: which numbers decide it and who owns them. A list without a decision is a junior answer.

Missed Corner

"Two lanes are simple: duplicate the block."

Packets from one flow may need to leave in order. The merge then needs sequence tags or a reorder buffer, which is new logic to verify.

Source Lesson P1.7 review claims (ordering, tags and out-of-order completion)

Learn It, Then Prove It