Testbench Best Practices

Professional verification patterns for SystemVerilog testbenches.

Layered Testbench Architecture

┌─────────────────────────────────────────┐
│              Test Layer                 │  ← Test scenarios
├─────────────────────────────────────────┤
│           Environment Layer             │  ← Agent coordination
├──────────────┬──────────────────────────┤
│    Agent     │    Agent                 │  ← Protocol-specific
│  ┌────────┐  │  ┌────────┐              │
│  │ Driver │  │  │Monitor │              │
│  └────────┘  │  └────────┘              │
│  ┌────────┐  │  ┌────────┐              │
│  │Sequencer│ │  │Scoreboard│            │
│  └────────┘  │  └────────┘              │
├──────────────┴──────────────────────────┤
│           Interface Layer               │  ← Signal abstraction
├─────────────────────────────────────────┤
│               DUT                       │
└─────────────────────────────────────────┘

---

1. Testbench Structure

Basic Self-Checking TB

module tb_dut;
    // Clock and reset
    logic clk = 0;
    logic rst_n = 0;
    always #5 clk = ~clk;

    // DUT signals
    logic [7:0] data_in, data_out;
    logic valid_in, valid_out;

    // DUT instance
    dut u_dut (.*);

    // Test
    initial begin
        $dumpfile("dump.vcd");
        $dumpvars(0, tb_dut);

        // Reset
        rst_n = 0;
        repeat(5) @(posedge clk);
        rst_n = 1;

        // Stimulus + Check
        for (int i = 0; i < 100; i++) begin
            @(posedge clk);
            data_in <= $urandom();
            valid_in <= 1;

            @(posedge clk);
            valid_in <= 0;

            // Wait for output
            wait(valid_out);
            check_result(data_in, data_out);
        end

        $display("TEST PASSED");
        $finish;
    end

    // Checker
    function void check_result(input [7:0] expected, input [7:0] actual);
        assert(actual == expected)
            else $error("Mismatch: expected %0h, got %0h", expected, actual);
    endfunction
endmodule

---

2. Transaction Class

Best Practice: Separate data from timing

class Transaction;
    rand bit [31:0] addr;
    rand bit [31:0] data;
    rand bit [3:0]  burst_len;
    rand bit        write;

    // Constraints
    constraint c_aligned { addr[1:0] == 2'b00; }
    constraint c_burst   { burst_len inside {1, 4, 8, 16}; }

    // Copy
    function Transaction copy();
        Transaction t = new();
        t.addr = this.addr;
        t.data = this.data;
        t.burst_len = this.burst_len;
        t.write = this.write;
        return t;
    endfunction

    // Display
    function void display(string prefix = "");
        $display("%s addr=%08h data=%08h burst=%0d %s",
            prefix, addr, data, burst_len, write ? "WR" : "RD");
    endfunction
endclass

---

3. Driver

Best Practice: Drive interface, not signals

class Driver;
    virtual bus_if.master vif;
    mailbox #(Transaction) mbx;

    function new(virtual bus_if.master vif, mailbox #(Transaction) mbx);
        this.vif = vif;
        this.mbx = mbx;
    endfunction

    task run();
        Transaction t;
        forever begin
            mbx.get(t);
            drive(t);
        end
    endtask

    task drive(Transaction t);
        @(posedge vif.clk);
        vif.addr  <= t.addr;
        vif.data  <= t.data;
        vif.valid <= 1;

        @(posedge vif.clk);
        while (!vif.ready) @(posedge vif.clk);
        vif.valid <= 0;
    endtask
endclass

---

4. Monitor

Best Practice: Passive observation only

class Monitor;
    virtual bus_if.monitor vif;
    mailbox #(Transaction) mbx;  // To scoreboard

    function new(virtual bus_if.monitor vif, mailbox #(Transaction) mbx);
        this.vif = vif;
        this.mbx = mbx;
    endfunction

    task run();
        Transaction t;
        forever begin
            @(posedge vif.clk);
            if (vif.valid && vif.ready) begin
                t = new();
                t.addr = vif.addr;
                t.data = vif.data;
                mbx.put(t);
            end
        end
    endtask
endclass

---

5. Scoreboard

Best Practice: Compare expected vs actual

class Scoreboard;
    mailbox #(Transaction) expected_mbx;
    mailbox #(Transaction) actual_mbx;
    int pass_count, fail_count;

    function new(mailbox #(Transaction) expected_mbx, mailbox #(Transaction) actual_mbx);
        this.expected_mbx = expected_mbx;
        this.actual_mbx = actual_mbx;
    endfunction

    task run();
        Transaction expected, actual;
        forever begin
            expected_mbx.get(expected);
            actual_mbx.get(actual);
            compare(expected, actual);
        end
    endtask

    function void compare(Transaction expected, Transaction actual);
        if (expected.data == actual.data) begin
            pass_count++;
        end else begin
            fail_count++;
            $error("MISMATCH: expected=%08h actual=%08h", expected.data, actual.data);
        end
    endfunction

    function void report();
        $display("=============================");
        $display("Scoreboard: PASS=%0d FAIL=%0d", pass_count, fail_count);
        $display("=============================");
    endfunction
endclass

---

6. Randomization Best Practices

Constraint Layering

class Transaction;
    rand bit [31:0] addr;
    rand bit [7:0] data;

    // Base constraint
    constraint c_valid_addr { addr < 32'h1000_0000; }

    // Can be overridden in extended class
    constraint c_mode { soft data > 0; }
endclass

class ErrorTransaction extends Transaction;
    // Override to inject errors
    constraint c_mode { data == 0; }
endclass

Weighted Distribution

class Packet;
    rand bit [1:0] pkt_type;

    constraint c_type_dist {
        pkt_type dist {
            0 := 60,  // 60% normal
            1 := 30,  // 30% error
            2 := 10   // 10% special
        };
    }
endclass

Solve Before

class Packet;
    rand bit [7:0] length;
    rand bit [7:0] payload[];

    constraint c_size {
        payload.size() == length;
        length inside {[1:64]};
        solve length before payload;
    }
endclass

---

7. Coverage Best Practices

Functional Coverage

class Coverage;
    Transaction t;

    covergroup cg_trans @(posedge clk);
        // Coverpoints
        cp_addr: coverpoint t.addr[31:28] {
            bins low  = {[0:3]};
            bins mid  = {[4:11]};
            bins high = {[12:15]};
        }

        cp_write: coverpoint t.write;

        cp_burst: coverpoint t.burst_len {
            bins single = {1};
            bins burst4 = {4};
            bins burst8 = {8};
            bins burst16 = {16};
            illegal_bins bad = default;
        }

        // Cross coverage
        cross_addr_write: cross cp_addr, cp_write;
    endgroup

    function new();
        cg_trans = new();
    endfunction

    function void sample(Transaction t);
        this.t = t;
        cg_trans.sample();
    endfunction
endclass

Coverage Goals

Coverage Type	Target
Line	>95%
Branch	>90%
FSM State	100%
FSM Transition	>95%
Functional	>98%

---

8. Threading Patterns

Fork/Join

// Parallel - wait for all
fork
    driver.run();
    monitor.run();
    scoreboard.run();
join

// Parallel - wait for any (with cleanup)
fork
    driver.run();
    monitor.run();
    timeout_check();
join_any
disable fork;

// Parallel - don't wait
fork
    background_task();
join_none

Timeout Pattern

task run_with_timeout(int cycles);
    fork
        begin
            run_test();
        end
        begin
            repeat(cycles) @(posedge clk);
            $error("TIMEOUT after %0d cycles", cycles);
            $finish;
        end
    join_any
    disable fork;
endtask

---

9. Interface Best Practices

Parameterized Interface

interface axi_if #(
    parameter int ADDR_W = 32,
    parameter int DATA_W = 32
) (
    input logic clk,
    input logic rst_n
);
    logic [ADDR_W-1:0] awaddr;
    logic [DATA_W-1:0] wdata;
    logic              awvalid, awready;
    logic              wvalid, wready;
    logic [1:0]        bresp;
    logic              bvalid, bready;

    modport master (
        output awaddr, awvalid, wdata, wvalid, bready,
        input  awready, wready, bresp, bvalid
    );

    modport slave (
        input  awaddr, awvalid, wdata, wvalid, bready,
        output awready, wready, bresp, bvalid
    );

    modport monitor (
        input awaddr, awvalid, awready, wdata, wvalid, wready,
              bresp, bvalid, bready
    );

    // Clocking block for TB
    clocking cb @(posedge clk);
        default input #1 output #1;
        output awaddr, awvalid, wdata, wvalid, bready;
        input  awready, wready, bresp, bvalid;
    endclocking
endinterface

Virtual Interface in Class

class Agent;
    virtual axi_if vif;

    function new(virtual axi_if vif);
        this.vif = vif;
    endfunction
endclass

---

10. Assertions in Testbench

Protocol Checks

// In interface or bind module
property p_valid_stable;
    @(posedge clk) disable iff (!rst_n)
    valid && !ready |=> valid && $stable(data);
endproperty
assert property (p_valid_stable) else $error("Valid dropped before ready");

property p_handshake;
    @(posedge clk) disable iff (!rst_n)
    valid |-> ##[1:100] ready;
endproperty
assert property (p_handshake) else $error("No ready within 100 cycles");

---

11. Test Organization

Test Base Class

virtual class BaseTest;
    Environment env;

    function new(Environment env);
        this.env = env;
    endfunction

    pure virtual task run();

    task pre_test();
        env.reset();
    endtask

    task post_test();
        env.report();
    endtask
endclass

class SmokeTest extends BaseTest;
    virtual task run();
        Transaction t = new();
        repeat(10) begin
            assert(t.randomize());
            env.driver_mbx.put(t);
        end
    endtask
endclass

---

12. Checklist

Before Starting

• Define verification plan
• Identify coverage goals
• List test scenarios
• Design TB architecture

During Development

• Use transactions, not raw signals
• Separate driver/monitor/scoreboard
• Use virtual interfaces
• Add functional coverage
• Include protocol assertions

Before Signoff

• All tests pass
• Coverage goals met
• No X/Z in simulation
• Edge cases tested
• Error injection tested

---

Reference

[SystemVerilog for Verification, 3rd ed. — Spear/Tumbush]
|url: https://picture.iczhiku.com/resource/eetop/wYIEDKFRorpoPvvV.pdf
|relevant: Ch 5 (OOP), Ch 6 (Rand), Ch 7 (Threads), Ch 8 (Advanced OOP), Ch 9 (Coverage), Ch 11 (Complete TB)

---

UVM-Lite for Verilator v5

Verilator v5 supports UVM 2017 subset. Works: uvm_component, uvm_object, uvm_sequence_item, uvm_driver, uvm_monitor, uvm_scoreboard, uvm_env, uvm_test, uvm_tlm ports, phases. Does NOT work: factory overrides, $cast, RAL, uvm_config_db with virtual interfaces, callbacks.

Pass virtual interfaces via constructor/setter, not config_db. Use direct construction, not factory. Compile with -DUVM_NO_DPI --timing.

Cocotb Equivalents

SV Pattern	Cocotb Equivalent
Transaction class	Python dataclass
rand/constraint	random module + manual constraints
mailbox	collections.deque or asyncio.Queue
virtual interface	Direct dut handle
covergroup	cocotb-coverage decorators
fork/join	Combine(start_soon(...))
fork/join_any	First(start_soon(...))
assert property	async def checker coroutine

Coverage Closure Checklist

1. Extract coverage report after full regression
2. Categorize: unreachable code (exclude with justification), untriggered FSM transitions (write directed test), empty cross bins (constraint override), low-hit coverpoints (increase iterations or add weighted distribution)
3. Write directed tests for each remaining hole
4. Re-run with merged coverage, verify all targets met
5. Document all exclusions with justification comments

10 TB Anti-Patterns

1. Magic numbers -> Use localparam with meaningful names
2. No timeout -> fork/join_any with $fatal timeout guard
3. Happy path only -> Test errors, resets mid-operation, edge cases
4. #delay for sync -> Use @(posedge clk) for synchronization
5. Ignoring X/Z -> Use $isunknown() and === operator
6. Monolithic test -> Split into reusable tasks
7. Print-and-pray -> Use assert with $error messages
8. Race conditions -> Non-blocking <= at clock edges
9. Global state -> Encapsulate in classes, pass via constructor
10. Not draining pipeline -> Wait PIPELINE_DEPTH + margin cycles before checking

Performance Optimization

Technique	Speedup
--threads N	2-4x
--trace-fst (not VCD)	2-3x smaller files
Disable tracing in regression	2-5x
Minimize $display	1.2-2x
--x-assign fast (after X-clean)	1.1x
Parallel test execution (make -j)	Nx

Test tiers: SMOKE (seconds, every commit), RANDOM (minutes, PR merge), COVERAGE (hours, nightly), REGRESSION (hours, weekly).