RTL design and micro architecture interviews test three things: whether you can describe hardware correctly in Verilog or SystemVerilog, whether you understand timing, clocking and reset well enough to build something that works on silicon, and whether you can reason about trade-offs in performance, power and area. The 50 questions below cover all three, from counters and FIFOs to pipelines, caches and clock domain crossing, and each one now comes with a short model answer.
The demand for capable RTL designers and micro architects remains strong, driven by the continuing push for higher efficiency, performance and functionality in semiconductor devices. Whether you are a fresh graduate or an experienced engineer refreshing your fundamentals, this list is meant to be worked through, not just read: try each answer aloud before reading the model answer.
Key takeaways
- RTL describes how data moves between registers on each clock edge; micro architecture decides which registers, pipelines, buffers and control exist in the first place.
- Most rejections come from basics: blocking versus non-blocking assignments, latches inferred by accident, unsafe clock domain crossings and resets.
- Know the setup and hold equations by heart and be able to put numbers into them.
- For every block you design (FIFO, arbiter, FSM), be ready to discuss corner cases: full and empty, back-pressure, reset values and what happens on the first and last cycle.
- Micro architecture answers should always end with a trade-off: what it costs in area, power, latency or verification effort.
What RTL design and micro architecture mean
RTL (register transfer level) design is the process of describing the flow of data and control between registers within a digital system. Micro architecture focuses on the detailed organisation of a processor or digital block one level above that: the pipeline depth, the queues and their sizes, the arbitration scheme, the cache organisation and the interfaces. Together they form the backbone of modern chip design. A good micro architecture specification makes the RTL almost mechanical to write; a weak one pushes design decisions into code where they are harder to review.
The 50 questions with model answers
1. What is RTL design and why is it important?
RTL design describes hardware as registers plus the combinational logic between them, written in an HDL such as Verilog, SystemVerilog or VHDL. It matters because it is the golden source for synthesis, simulation, formal checks and equivalence checking. Every later step, from gate netlist to layout, is checked against it.
2. Describe the difference between combinational and sequential logic.
Combinational logic outputs depend only on current inputs (adders, multiplexers, decoders). Sequential logic has memory: outputs depend on current inputs and stored state, held in flip-flops or latches (counters, FSMs, shift registers).
3. What is the difference between synchronous and asynchronous designs?
In a synchronous design, state changes only on a common clock edge, which makes timing analysis by static timing analysis (STA) straightforward. Asynchronous designs use handshakes or self-timed logic instead of a global clock. In interviews the question often also means synchronous versus asynchronous reset: a synchronous reset is sampled on the clock edge, an asynchronous reset acts immediately but must be de-asserted synchronously to avoid metastability.
4. Explain the concept of pipelining in microarchitecture.
Pipelining splits a long combinational path into stages separated by registers, so a new operation can start every cycle. Illustrative example: a 10 ns block split into five 2 ns stages, with 0.2 ns of register overhead per stage, runs at a 2.2 ns clock. Throughput rises about 4.5 times, but latency grows from 10 ns to 11 ns, and you pay for the extra flops and for hazard handling.
5. Describe the basic structure of a Verilog module.
A module has a name, a port list with directions and widths, optional parameters, internal declarations (wires, regs), and a body of continuous assignments, always blocks and instances of other modules, closed by endmodule. The counter below shows all the usual parts.
6. How do you implement a simple counter in Verilog?
// 4-bit up counter with synchronous enable, asynchronous active-low reset
module counter4 (
input wire clk,
input wire rst_n,
input wire en,
output reg [3:0] count
);
always @(posedge clk or negedge rst_n) begin
if (!rst_n)
count <= 4'd0;
else if (en)
count <= count + 4'd1;
end
endmodule
Expect follow-ups on wrap-around, a load input, a terminal-count output and how the reset style affects synthesis.
7. Explain the difference between blocking and non-blocking assignments in Verilog.
Blocking (=) updates the variable immediately, so the next statement sees the new value. Non-blocking (<=) schedules the update for the end of the time step, so all right-hand sides are read first. Use non-blocking for sequential logic and blocking for combinational logic. Mixing them in clocked blocks causes simulation and synthesis mismatches and race conditions; Clifford Cummings' SNUG 2000 paper on non-blocking assignments is the standard reference.
8. What is a state machine and how do you implement it in RTL?
A finite state machine (FSM) has a set of states, a state register, next-state logic and output logic. The usual coding style is a clocked block for the state register and a combinational block with a case statement for next state, with a default assignment at the top so no latch is inferred. Discuss state encoding (binary, one-hot, Gray) and what happens on an illegal state.
9. Describe the concept of a clock domain and how you handle clock domain crossing.
A clock domain is all logic driven by one clock (or clocks with a fixed phase relationship). Crossing between unrelated clocks needs a synchroniser: two flip-flops for a single-bit level signal, a handshake or pulse synchroniser for events, and an asynchronous FIFO with Gray-coded pointers for multi-bit data. Never synchronise a multi-bit bus bit by bit, because bits can resolve on different cycles.
// Two-flop synchroniser for a single-bit level signal
module sync2 (input wire clk_dst, input wire rst_n,
input wire d_async, output wire q_sync);
reg meta, stable;
always @(posedge clk_dst or negedge rst_n) begin
if (!rst_n) begin
meta <= 1'b0;
stable <= 1'b0;
end else begin
meta <= d_async;
stable <= meta;
end
end
assign q_sync = stable;
endmodule
10. What is the significance of setup and hold times in digital design?
Setup time is how long data must be stable before the capturing clock edge; hold time is how long it must stay stable after. With skew defined as capture clock arrival minus launch clock arrival:
- Setup: Tclk >= tcq,max + tlogic,max + tsetup minus skew
- Hold: tcq,min + tlogic,min >= thold + skew
Illustrative numbers: tcq 80 ps, logic 700 ps, setup 50 ps and skew +20 ps give a minimum period of 810 ps, about 1.23 GHz. Setup failures can be fixed by slowing the clock; hold failures cannot, because hold does not depend on the period.
11. Explain the role of metastability in digital circuits.
If data changes inside the setup and hold window, a flip-flop can sit at an intermediate voltage for an unbounded time before resolving. It cannot be prevented, only made rare. The usual estimate is MTBF = etr/τ / (T0 · fclk · fdata), where tr is the time allowed to resolve and τ is a technology constant. Because the dependence is exponential, an extra synchroniser stage that adds a full clock period of resolution time improves MTBF enormously. Illustratively, with τ = 50 ps, adding 1 ns of resolution time multiplies MTBF by e20, about 5 x 108.
12. Describe how you would implement a FIFO (First-In-First-Out) buffer in RTL.
A synchronous FIFO is a memory with a write pointer and a read pointer. Adding one extra bit to each pointer separates full from empty: equal pointers mean empty; equal address bits with different top bits mean full.
// Synchronous FIFO, depth = 2**AW
module sync_fifo #(parameter WIDTH = 8, parameter AW = 2) (
input wire clk,
input wire rst_n,
input wire wr_en,
input wire [WIDTH-1:0] wdata,
input wire rd_en,
output wire [WIDTH-1:0] rdata,
output wire full,
output wire empty
);
localparam DEPTH = 1 << AW;
reg [WIDTH-1:0] mem [0:DEPTH-1];
reg [AW:0] wptr, rptr; // one extra bit tells full from empty
wire do_wr = wr_en && !full;
wire do_rd = rd_en && !empty;
assign empty = (wptr == rptr);
assign full = (wptr[AW] != rptr[AW]) && (wptr[AW-1:0] == rptr[AW-1:0]);
assign rdata = mem[rptr[AW-1:0]]; // first-word fall-through read
always @(posedge clk) if (do_wr) mem[wptr[AW-1:0]] <= wdata;
always @(posedge clk or negedge rst_n) begin
if (!rst_n) begin
wptr <= 0;
rptr <= 0;
end else begin
if (do_wr) wptr <= wptr + 1'b1;
if (do_rd) rptr <= rptr + 1'b1;
end
end
endmodule
For an asynchronous FIFO, the pointers are converted to Gray code and passed through two-flop synchronisers into the other domain, as described in Cummings' SNUG 2002 paper on asynchronous FIFO design. Interviewers will also ask how to size the depth from burst length and the two clock rates.
13. What is clock gating and why is it used?
Clock gating stops the clock to registers that are not changing, saving dynamic power in the clock tree and the flops. It is done with an integrated clock gating (ICG) cell, a latch plus an AND gate, so the enable cannot glitch the gated clock. A plain AND gate on a clock is a classic interview trap.
14. Explain the concept of an ALU (Arithmetic Logic Unit) and its design.
An ALU performs arithmetic (add, subtract, compare) and logic (AND, OR, XOR, shifts) operations selected by an opcode, and usually produces flags such as zero, carry, negative and overflow. Design questions focus on sharing one adder for add, subtract and compare, and on the critical path through the adder and result multiplexer.
15. How do you design a register file?
A register file is a small, fast array with a fixed number of read and write ports, for example two reads and one write per cycle for a simple RISC core. Small ones are built from flops, larger ones from custom or compiled memories. Discuss write-then-read bypass in the same cycle, register 0 hard-wired to zero in RISC-V, and how port count grows area roughly with the square of the ports.
16. Describe the process of designing a memory controller.
Start from the memory standard (for DRAM, the JEDEC DDR or LPDDR specification) and the traffic profile. The controller accepts requests, queues and reorders them to keep rows open and banks busy, issues activate, read, write and precharge commands within timing parameters such as tRCD, tRP and tRAS, schedules refresh, and handles the PHY interface. Trade-offs are latency versus bandwidth and fairness versus efficiency.
17. What are the differences between Mealy and Moore state machines?
Moore outputs depend only on the current state, so they change only after a clock edge and are glitch-free if registered. Mealy outputs depend on state and inputs, so they react in the same cycle and often need fewer states, but an input glitch can reach the output combinationally.
18. Explain the concept of a pipeline hazard and how to resolve it.
A hazard is a situation where the next instruction cannot execute in its planned cycle. Structural hazards (two stages need one resource) are solved by duplicating resources; data hazards by forwarding or stalling; control hazards by branch prediction, early branch resolution or flushing.
19. How do you handle data forwarding in a pipelined processor?
Results from the EX/MEM and MEM/WB pipeline registers are fed back through multiplexers to the ALU inputs when the destination register of an older instruction matches a source register of a younger one. A load followed by a dependent instruction still needs a one-cycle stall in a classic five-stage pipeline, because the data is only available after the memory stage.
20. Describe the differences between RISC and CISC architectures.
RISC uses a small set of fixed-length, simple instructions with load/store memory access, which suits pipelining. CISC has variable-length instructions that can operate on memory directly. Modern x86 processors decode CISC instructions into internal micro-operations, so the difference today lies more in the instruction set and decoder than in the execution core.
21. What is a branch predictor and how does it improve processor performance?
A branch predictor guesses the direction and target of a branch before it resolves, so fetch can continue without bubbles. Simple schemes use 2-bit saturating counters; advanced ones use global history (gshare, TAGE) and a branch target buffer. Each misprediction costs roughly the pipeline depth to the resolve stage, so accuracy matters more as pipelines get deeper.
22. Explain the concept of superscalar architecture.
A superscalar processor issues more than one instruction per cycle using multiple execution units. It needs wider fetch and decode, dependency checking, more register file ports and often out-of-order execution with register renaming and a reorder buffer.
23. Describe how to design a cache memory and its various types (L1, L2, L3).
A cache is defined by size, line size, associativity, write policy (write-through or write-back) and replacement policy. L1 is small and fast, usually split into instruction and data caches; L2 is larger and slower; L3 is often shared between cores. A useful worked figure is average memory access time: AMAT = hit time + miss rate x miss penalty. Illustratively, 1 cycle + 0.05 x 100 cycles = 6 cycles, so halving the miss rate saves more than shaving the hit time.
24. What are the common techniques for cache replacement policies?
Least recently used (LRU), pseudo-LRU (tree-based, cheaper in hardware), first in first out, random, and newer schemes such as RRIP that resist scanning access patterns. True LRU becomes expensive above four to eight ways.
25. How do you implement a 4-bit adder in Verilog?
Behaviourally, assign {cout, sum} = a + b + cin; lets synthesis choose the architecture. Structurally, a ripple-carry adder chains full adders:
module full_adder (input wire a, b, cin, output wire s, cout);
assign s = a ^ b ^ cin;
assign cout = (a & b) | (cin & (a ^ b));
endmodule
module adder4 (
input wire [3:0] a, b,
input wire cin,
output wire [3:0] sum,
output wire cout
);
wire [4:0] c;
assign c[0] = cin;
genvar i;
generate
for (i = 0; i < 4; i = i + 1) begin : g_fa
full_adder u_fa (.a(a[i]), .b(b[i]), .cin(c[i]), .s(sum[i]), .cout(c[i+1]));
end
endgenerate
assign cout = c[4];
endmodule
Follow-ups: why ripple carry is slow for wide adders, and how carry-lookahead or carry-select adders shorten the path.
26. Explain the concept of instruction pipelining and its benefits.
Instruction pipelining overlaps fetch, decode, execute, memory and write-back of successive instructions. Ideally one instruction completes per cycle, raising throughput and allowing a higher clock, at the cost of hazards, more registers and more complex control.
27. Describe the role of a control unit in a processor.
The control unit decodes instructions and generates the signals that steer the datapath: multiplexer selects, ALU operation, register write enables, memory read and write, and stall and flush controls. It can be hard-wired logic or microcoded.
28. What is an interlock and how is it used in processor design?
An interlock is hardware that detects a hazard and stalls the pipeline (inserting a bubble) until it is safe to continue, for example on a load-use dependency. Early MIPS designs avoided interlocks with delay slots; most modern designs interlock in hardware.
29. Explain the concept of speculative execution.
The processor executes instructions before it knows they are needed, usually past a predicted branch, and discards the results if the guess was wrong. A reorder buffer keeps results uncommitted until they are known to be correct. Spectre and Meltdown (2018) showed that speculation can leak data through cache side channels, so security now features in these discussions.
30. Describe how you would design an interrupt controller.
Capture interrupt sources (edge or level), latch pending bits, apply masks and priorities, present the highest-priority vector to the core, and clear on acknowledge or end of interrupt. Discuss synchronising external interrupt inputs, nesting and priority preemption, and how software reads and clears status registers. The Arm GIC and the RISC-V PLIC are good reference designs.
31. What is the significance of power gating in digital design?
Power gating switches off the supply to idle blocks with header or footer switch transistors, cutting leakage. It needs isolation cells on outputs, retention flops if state must survive, controlled power-up sequencing to limit inrush current, and a power intent description, usually in UPF (IEEE 1801).
32. How do you ensure timing closure in a complex design?
Begin at micro architecture with realistic pipeline depth and register boundaries at block outputs. Write clean constraints (clocks, generated clocks, false and multicycle paths reviewed by the designer), run synthesis and STA early, fix long paths in RTL by retiming or restructuring, and track timing across corners and modes. Most hard timing problems are architecture problems discovered late.
33. Describe the process of designing a multiplier in RTL.
Options range from a sequential shift-and-add multiplier (small, many cycles) to array multipliers and Booth-encoded Wallace or Dadda trees (fast, large). In practice a * b is usually left to synthesis or a DesignWare-style library component, pipelined to meet frequency. Interviewers ask about signed operands and pipeline depth.
34. What is the role of a testbench in digital design verification?
A testbench generates stimulus, drives the design under test, checks outputs against expected results and measures coverage. Modern testbenches use SystemVerilog and UVM with constrained random stimulus, scoreboards, assertions and functional coverage.
35. Explain the concept of data hazards and how to mitigate them.
Data hazards are read after write (true dependency), write after read and write after write (name dependencies). RAW hazards are handled by forwarding and stalls; WAR and WAW matter in out-of-order machines and are removed by register renaming.
36. Describe the process of designing a DMA (Direct Memory Access) controller.
A DMA controller moves data between memory and peripherals without the CPU. Define the descriptor format (source, destination, length, control), channels and their arbitration, bus interface (for example AXI with burst support), alignment handling, completion interrupts and error reporting. Scatter-gather uses linked descriptors.
37. What are the key differences between SRAM and DRAM?
| Property | SRAM | DRAM |
|---|---|---|
| Typical cell | 6 transistors (6T) | 1 transistor, 1 capacitor |
| Refresh | Not needed | Needed, charge leaks |
| Speed | Faster | Slower |
| Density and cost per bit | Lower density, higher cost | Higher density, lower cost |
| Typical use | Caches, register files, on-chip buffers | Main memory |
38. Explain the concept of bus arbitration and its importance.
When several masters share a bus or resource, an arbiter decides who wins each cycle. Fixed priority is simple but can starve low-priority masters; round-robin is fair; weighted schemes balance bandwidth and latency. Ask yourself how the arbiter behaves when the winner is back-pressured.
39. How do you design a low-power digital circuit?
Dynamic power is roughly αCV2f, so reduce switching activity (clock gating, operand isolation, avoiding glitches), capacitance and voltage (multiple voltage domains, dynamic voltage and frequency scaling). Reduce leakage with power gating and high-threshold cells off the critical path. The biggest wins come from architecture, such as not doing work that is not needed.
40. Describe the differences between level-sensitive and edge-triggered flip-flops.
Strictly, a level-sensitive storage element is a latch: it is transparent while the enable is active. An edge-triggered flip-flop samples only on the clock edge. Latches allow time borrowing and are smaller, but make timing analysis harder; flip-flops are the default in most RTL. In code, an incomplete if or case in a combinational block infers an unwanted latch.
41. Explain the concept of a sequencer in microarchitecture.
A sequencer generates the order of control steps, for example the next microinstruction address in a microcoded control unit, or the sequence of operations in a multi-cycle engine. It is essentially an FSM or a counter with branch logic. (In UVM, "sequencer" also names the component that feeds sequences to a driver.)
42. How do you implement a priority encoder in Verilog?
// 8-to-3 priority encoder, bit 7 has the highest priority
module prio_enc8 (
input wire [7:0] req,
output reg [2:0] idx,
output wire valid
);
integer k;
assign valid = |req;
always @(*) begin
idx = 3'd0;
for (k = 0; k < 8; k = k + 1)
if (req[k]) idx = k[2:0]; // higher bits overwrite lower ones
end
endmodule
The default assignment at the top prevents latch inference, and the valid output tells "request 0" apart from "no request".
43. What are the common techniques for clock skew management?
Balanced clock tree synthesis, clock meshes or spines for high-performance designs, matching buffer types and loads, and useful skew where the tool deliberately skews clocks to borrow time. From RTL, keep clock structures simple: few gating levels, no logic on clocks except ICGs and approved multiplexers.
44. Describe the process of designing a digital filter in RTL.
Start from the filter specification and a fixed-point model (often in MATLAB or Python). Choose FIR or IIR, decide coefficient and data bit widths, then map multiply-accumulate operations into a direct, transposed or systolic structure, pipelined for frequency. Verify against the model bit for bit, and handle rounding, saturation and overflow explicitly.
45. Explain the concept of hardware description languages (HDLs).
HDLs such as Verilog (IEEE 1364), SystemVerilog (IEEE 1800) and VHDL (IEEE 1076) describe concurrent hardware behaviour and structure. Unlike software, every always block and continuous assignment runs in parallel, and only a synthesizable subset maps to gates.
46. What is a latch-up and how can it be prevented in CMOS circuits?
Latch-up is the triggering of a parasitic PNPN (thyristor) structure formed by the n-well, p-substrate and the transistor diffusions, which creates a low-resistance path from supply to ground and can destroy the chip. Prevention includes frequent well and substrate taps, guard rings around I/O devices, adequate spacing, and process options such as epitaxial wafers or SOI. JEDEC JESD78 defines the latch-up test. This is a circuit and layout topic, so interviewers mainly check that you know it is not an RTL bug.
47. How do you design a high-speed serializer/deserializer (SerDes) interface?
The analog PHY handles the transmitter, equaliser and clock and data recovery. The digital side, which RTL designers own, covers the physical coding sublayer: encoding (8b/10b or 64b/66b), scrambling, word alignment, elastic buffers for clock compensation, lane deskew, PRBS generators and checkers for test, and the width conversion between the parallel core clock and the PHY interface.
48. Describe the process of creating a microarchitecture specification document.
Start from the requirements, then describe the block diagram, interfaces and protocols, data flow, pipeline and timing diagrams, FSMs, register map, clocks and resets, power modes, performance calculations, error handling, and DFT and verification hooks. A good specification states why each choice was made and what alternatives were rejected.
49. What is the role of a pipeline register in a processor pipeline?
A pipeline register holds the data and control of one instruction between stages, isolating stages so each has a full cycle. It also carries valid bits, and supports stalls (hold its value) and flushes (clear the valid bit).
50. Explain the concept of hardware/software co-design.
Hardware/software co-design partitions a system's functions between hardware and software early, using profiling and models to decide what deserves an accelerator. It covers the interface between them: registers, interrupts, DMA, drivers and firmware. Virtual platforms and FPGA prototypes let software start before silicon.
How interviewers probe these topics
- Write it on the board: a counter, FIFO, edge detector or arbiter, then "now make it parameterised" or "now add back-pressure".
- Find the bug: code with a missing default in a
case, a blocking assignment in a clocked block, or a multi-bit bus crossing clock domains. - Put numbers in: compute the maximum frequency from given delays, or a FIFO depth from burst size and clock rates.
- Change one constraint: "the clock doubles" or "the memory now has two-cycle latency"; they want to see how your micro architecture adapts.
- Trade-offs: for any answer, expect "what does that cost?"
Common mistakes candidates make
- Saying hold violations can be fixed by lowering the clock frequency.
- Using a plain AND gate for clock gating.
- Synchronising each bit of a counter separately across clock domains instead of using Gray code or a handshake.
- Forgetting reset values, or using asynchronous reset de-assertion without a reset synchroniser.
- Describing a design without giving latency, throughput or area numbers.
FAQ
Which language should I prepare in, Verilog or SystemVerilog?
Write RTL in whatever the team uses, but SystemVerilog is now common for both design and verification. Knowing Verilog well, and the SystemVerilog additions such as always_ff, always_comb, logic and interfaces, covers most interviews.
How deep do micro architecture questions go for a fresh graduate?
Expect pipelining, hazards, caches and FSMs at textbook level, plus one small design problem. Experienced candidates are asked to defend real design decisions from their past projects.
Do RTL designers need to know STA?
Yes, at least setup and hold, clock definitions, false and multicycle paths, and how RTL structure affects the critical path. Timing problems fixed in RTL are cheaper than those fixed in layout.
What is the best way to practise?
Write and simulate the small blocks above with a free simulator such as Icarus Verilog or Verilator, then explain each design aloud, including its corner cases and costs.


.jpeg)
Your comments will be moderated before it appears here.