HDL and verification basics
Introduit| English | Sens |
|---|---|
| simulation/ˌsɪmjʊˈleɪʃn/ | Running a model of a circuit on a computer to predict how it will behave. |
| hardware description language/ˈhɑːdweə dɪˈskrɪpʃn ˈlæŋɡwɪdʒ/ | A language such as Verilog or VHDL used to describe and simulate digital circuits. |
| synthesizable/ˈsɪnθəsaɪzəbl/ | synthétisable |
| testbench/ˈtestbentʃ/ | banc de test |
64 gates, 64 chances to slip
- Wire 64 gates by hand and each connection is a chance for a mistake. Modern chips have billions.
- Industry does not answer this with more careful drawing. It describes the hardware in a language and proves the description with simulation.
Describe hardware in code
- A hardware description language 硬件描述语言 (HDL) such as Verilog or VHDL describes what a circuit is and how it behaves. A synthesis tool turns it into gates.
- Only synthesizable 可综合 code can become hardware: continuous
assign,always @(*)blocks andalways @(posedge clk)registers.
Any Verilog that simulates correctly can be turned into gates.
Difficulté 1/5 · 100 pièces + un petit bonus lors de la première réussite aujourd'hui
Simulation accepts constructs with no hardware meaning, such as delays and $display. Only the synthesizable subset becomes gates.
A full adder in four lines
- The module lists its ports, then states the two output equations from the earlier lesson.
module full_adder(input a, b, cin,
output sum, cout);
assign sum = a ^ b ^ cin;
assign cout = (a & b) | (cin & (a ^ b));
endmodule
^is XOR,&is AND,|is OR. Everyassignruns all the time, in parallel, like the wires it stands for.
Match each Verilog operator to its gate.
Difficulté 2/5 · 140 pièces + un petit bonus lors de la première réussite aujourd'hui
These bitwise operators map one-to-one onto gates, which is why assign lines read like a schematic.
The testbench checks it
- A testbench 测试平台 is code that drives inputs into the module and checks the outputs. It runs only in simulation 仿真 and never becomes hardware.
- It may use things hardware cannot: delays such as
#10, loops over test cases, and$displayto print results.
Which construct belongs ONLY in a testbench?
Difficulté 1/5 · 100 pièces + un petit bonus lors de la première réussite aujourd'hui
Delays simulate but do not synthesise: hardware has no #10. The other three all become gates or flip-flops.
Worked example: an independent check
- The testbench drives all 8 combinations of
a,b,cinand checks $a + b + c_{in} = 2 \cdot c_{out} + s$ each time. All 8 pass. - The check uses arithmetic, not a second gate description. A copy of the design would repeat any mistake in it, so it would prove nothing.
An exhaustive testbench for a block with 5 one-bit inputs must try how many input combinations?
Difficulté 2/5 · 140 pièces + un petit bonus lors de la première réussite aujourd'hui
$2^{5} = 32$. Small blocks are cheap to check completely; a 64-bit adder is not, which is why big designs mix exhaustive checks of parts with random tests.
A testbench that checks a full adder against a second gate copy of the same design is —
Difficulté 1/5 · 100 pièces + un petit bonus lors de la première réussite aujourd'hui
The reference must be independent: arithmetic ($a + b + c_{in} = 2c_{out} + s$) or a hand trace. A restated implementation repeats the very bug you are looking for.
Worked example: a shift register
- A 4-bit serial-in shift register is one clocked line. The concatenation
{q[2:0], sin}drops the top bit and putssinin at the right.
module shift4(input clk, sin,
output reg [3:0] q);
always @(posedge clk)
q <= {q[2:0], sin};
endmodule
- From 0000 with
sin= 1, 0, 1, 1, simulation shows q = 0001, 0010, 0101, 1011.
The 4-bit shift register q <= {q[2:0], sin}, starting at 0000, with sin = 1, 0, 1, 1 over four clocks traces q as —
Difficulté 1/5 · 100 pièces + un petit bonus lors de la première réussite aujourd'hui
Each edge shifts left and admits sin on the right, giving 0001, 0010, 0101, 1011. The concatenation is the shift.
Three common slips
- A testbench that restates the design. It checks the code against itself; use arithmetic or a hand trace.
- Testbench constructs in a hardware module. Delays and
$displaydo not synthesise. - No reset. A register without one starts in an unknown state in hardware, even if simulation looks fine.
A counter module has no reset. In simulation every register starts at x (unknown). What does that tell you?
Difficulté 1/5 · 100 pièces + un petit bonus lors de la première réussite aujourd'hui
Real flip-flops power up at random values. The x in simulation is a warning, not a simulator fault; a reset puts the counter in a known state.
Recap
- Synthesizable modules become gates; testbenches only simulate.
- Check against an independent reference: arithmetic or a traced sequence.
- Small input spaces are checked exhaustively; ship each module with its testbench.