About

Log in?

DTU users get better search results including licensed content and discounts on order fees.

Anyone can log in and get personalized features such as favorites, tags and feeds.

Log in as DTU user Log in as non-DTU user No thanks

DTU Findit

Report

Clock domain crossing modules for OCP-style read/write interfaces

From

Technical University of Denmark1

Department of Applied Mathematics and Computer Science, Technical University of Denmark2

Embedded Systems Engineering, Department of Applied Mathematics and Computer Science, Technical University of Denmark3

The open core protocol (OCP) is an openly licensed, configurable, and scalable interface protocol for on-chip subsystem communications. The protocol defines read and write transactions from a master towards a slave across a point-to-point connection and the protocol assumes a single common clock. This paper presents the design of two OCP clock domain crossing interface modules, that can be used to construct systems with multiple clock domains.

One module (called OCPio) supports a single word read-write interface and the other module (called OCPburst) supports a four word burst read-write interface. The modules has been developed for the T-CREST multi-core platform [8, 9, 13] but they can easily be adopted and used in other designs implementing variants of the OCP interface standard.

The OCPio module is used to connect a Patmos processor to a message passing network-on-chip and the OCPburst is used to connect the Processor and its cache controllers to a shared o_-chip memory. While the problem of synchronizing a simple streaming interface is well described in the literature and often solved using bi-synchronous FIFOs we found surprisingly little published material addressing synchronization of bus-style read-write transaction interfaces.

An OCP interface typically has control signals related to both the master issuing a read or write request and the slave producing a response. If all these control signals are passed across the clock domain boundary and synchronized it may add significant latency to the duration of a transaction. Our interface designs avoid this and synchronize only a single signal transition in each direction during a read or a write transaction.

The designs are available as open source, and the modules have been tested in a complete multi-core platform implemented on an FPGA board.

Language: English
Publisher: Technical University of Denmark
Year: 2016
Series: Dtu Compute Technical Report-2016
ISSN: 16012321
Types: Report
ORCIDs: Sparsø, Jens

DTU users get better search results including licensed content and discounts on order fees.

Log in as DTU user

Access

Analysis