Transaction level modelling ?

Transaction-level models (TLMs) are architecture models that exchange transactions, i.e. abstract data structures, instead of signals. This means that instead of having to toggle hundreds of signals to perform a data write on a bus, this operation is encapsulated in a data structure and communicated to an abstract representation (TLM) of the bus, hence completing the transaction a lot faster, but sacrificing at the same time a certain degree of accuracy.

TLMs are mostly used for rapid prototyping, performance analysis and could serve as "golden" models for your RTL verification. There are usually several abstraction levels for implementing a TLM according to your needs, such as un-timed models, cycle-accurate and even pin-accurate models. The more elaborate the abstraction layer, the more simulation time it will consume, so the performance can end up being comparable to RTL simulation.

To implement TLM, it is usually not a question of language, as most modern languages support this, but a question of methodology. Ideally, you would want to transition from the TLM environment to a RTL environment, simply by adding BFMs and monitors that translate the transactions from/to signals, while keeping the rest of the environment intact. uRM offers a lot of information on how to achieve this, even in mixed-language environments.

Hi Iraklis.

thanks for sharing the info. can you please add one example if possible.


-Manmohan

Hi Manmohan,

I am not aware of a SystemVerilog TLM example, but the Incisive Plan to Closure Methodology (IPCM) offers a TLM example in SystemC, which you might find interesting as a starting point. Apart from useful documentation, you will also find extensive code examples that highlight the TLM principles. If possible, I strongly recommend you to consider SystemC as a modeling language, as it was designed purely for that purpose.

If anyone is aware of a SystemVerilog TLM example, please let us know.

Hi...........

Can we write TLM in 'e' language?
Or we can write only signal level model in 'e' using BFM.

Thanks!

Hello,

You can build a transaction-level model in any high-level structured programming language, including e, SystemVerilog, SystemC, C/C++ and more.

A transaction level model is usually a high-speed model that trade-offs speed for cycle accuracy to analyze system-level performance and features. Therefore there isn't a need to build signal level drivers, monitors. The components exchange data using a function call passing objects and synchronizing on high-level system events.

Users have built transaction level models in e for years mainly for verification (separating data generation, scoreboaring, coverage, from pin-level drivers monitors) as well as for modeling systems. This is the methodology that eRM and now OVM recommends for structuring a verification environment for modularity and reuse.

In OVM, there is also a TLM api provided to assist with data exchange between components using well defined semantics. For example, blocking/non-blocking requests of data, 0-time broadcast of data, FIFO, etc.

You can find a couple of TLM examples in OVM distribution in ovm-1.1/examples/tlm. You can get this distribution at http://www.ovmworld.org/downloads.php.

Thanks,

Umer