Multiprocessor Architecture Watch
Concurrency Abstraction Watch

Memory Architecture View

The Memory View aims to model a large memory design space at a high level of abstraction. The memory design space ranges from simple memory elements, such as register files, to complex memory systems, such as memory hierarchies. Designers model memory architectures in the Memory View by assembling and configuring components from an extendable library. There are two main benefits of the Memory View. First, the parameterized library elements enable automated exploration during the initial design stage. Second, the flexibility allows an expert memory architect to explore novel designs without starting from the scratch.

Memory architectures described in the Memory View are formal, functional models. All memory elements and systems are built bottom up by configuring and connecting three basic actors: table actors, access actors and arbitrate actors. The table actor is used as a storage element. The access actor is used to access table actors for storing and retrieving data. The arbitrate actor is used to order multiple accesses by different access actors into the same table actor. The Memory View's underlying formalism guarantees precise descriptions of the memory system and enables a correct by construction synthesis path towards an implementation. To ensure consistency with other views, an interface is used to connect the Memory View to the Processing Element Architecture View. This interface allows a memory system described in the Memory View to be used within a Processing Element model as a Processing Element actor. The functional consistency between the two views is preserved.

Send feedback to mescal@gigascale.org
You are not logged in 
©2002-2018 U.C. Regents