Pausible and Stoppable Clocks in GALS

Traditionally, clocks are derived from very accurate, crystal-controlled oscillators whose frequencies are fixed. Any interface between two independently clocked domains has to assume that the timing of the two clocks cannot be changed in any way, and this leads to the need for synchronizers to retime the data. In a globally asynchronous, locally synchronous (GALS) system [65,67,68], it is possible to pause, or stop the clock, rather than retime the data so that synchronization is not necessary.

There are two main methods of using the communication between processors to link the clocks. If the processor clock starts when data arrives, and activity stops as soon as the data is processed the clock is called data driven [71]. With a data-driven clock there is no choice between ongoing internal processor activity and the need to process new data, so there is no competition between them and there can be no metastable signals. A pausible clock [72,73] is normally running when the input data arrives so there is still a choice, to pause or not to pause, just before the next clock is due. Even if there is no internal processing when data arrives, requests coming from two independent sources to the same processor may have to be arbitrated. Instead of synchronizers with a fixed resolution time and a consequent probability of failure, a processor core with a pausible clock uses a MUTEX to decide if the clock should be paused in this cycle or the next. ...

Get Synchronization and Arbitration in Digital Systems now with the O’Reilly learning platform.

O’Reilly members experience live online training, plus books, videos, and digital content from nearly 200 publishers.