Transaction-Level Modeling (TLM)

Components talk in transactions, not wires

TLM lets components exchange whole transactions (objects) through standardized interfaces instead of poking each other's internals. This decoupling is what lets a monitor, scoreboard, and coverage collector be written independently and still snap together in connect_phase.

TLM topic map

Legend: [UVM] [TLM] [MON] [CHECK]

┌─────────────────────────────────────────────────────────────────────────┐
│  TRANSACTION-LEVEL MODELING — topic map (each hub has 5–6 sub-lessons)  │
├─────────────────────────────────────────────────────────────────────────┤
│  1. PORT MODEL [TLM]   port │ export │ imp │ connect │ hierarchy pass   │
│  2. PUT/GET [TLM]       blocking │ non-blocking │ transport │ flow ctrl   │
│  3. ANALYSIS [MON]      analysis_port │ subscriber │ write broadcast      │
│  4. FIFOs [TLM]         tlm_fifo │ analysis_fifo │ producer/consumer      │
│  5. DEBUG [UVM]         unconnected │ type mismatch │ silent drops         │
└─────────────────────────────────────────────────────────────────────────┘

Two TLM styles

POINT-TO-POINT (flow controlled)        BROADCAST (fire-and-forget)
  port  -> export -> imp                   analysis_port -> imp(s)
  put / get / peek / transport             write() to many subscribers
  one initiator, one target                one initiator, N targets
  driver<->sequencer, producer<->consumer  monitor -> scoreboard + coverage

Topics and sub-topics

1. Ports, Exports & Imps — 6 sub-lessons on initiator/target roles, connect rules, and hierarchical pass-through.

2. put / get / transport — 6 sub-lessons on blocking vs non-blocking and transport combined ops.

3. Analysis Ports & uvm_subscriber — 6 sub-lessons on broadcast write(), subscribers, and monitor wiring.

4. TLM FIFOs & Analysis FIFO — 6 sub-lessons on decoupling threads and FIFO patterns.

5. TLM Debugging — 6 sub-lessons on unconnected ports, type mismatches, and silent drops.

Key takeaways

• TLM moves transactions, not pins — components stay decoupled and reusable.

• Point-to-point (put/get) has flow control; analysis (write) is broadcast with none.

• The monitor → scoreboard/coverage path is the canonical analysis broadcast pattern.

How to use this section

Open a topic hub for an overview and diagram, then drill into sub-lessons. When wiring envs, always verify TLM connections in connect_phase — unconnected ports are the #1 silent-failure source.

• Initiator calls through a port; target implements via imp; export bridges hierarchy levels.

• Monitors use analysis_port.write() — never put/get for passive observation.

• Use uvm_tlm_if print or connection debug when transactions mysteriously disappear.