|
Network-on-Chip
Credits: 1 ECTS
Description The aim
of this module is to provide an overview of the network-on-chip (NoC)
concepts and highlighting on-going research in the design of such a network.
The NoC overview, presented here, is based on abstraction, namely: the
system, the network-adapter, the network, and the link layers.
- The system layer deals with modeling and simulation of network's
traffic source and sink i.e. the application mapped on a homogeneous or
heterogeneous system-on-chip architectures. We have classified the traffic
generated by the IP cores into latency-critical, streaming and best-effort
classification.
- The network-adapter layers describes encapsulation and socket issues
at the interface between the network and the IP cores. The important
issues at the network and link layer, covered in the course, are: (i)
topology, (ii) protocol, (iii) flow control, and (iv) quality-of-service (QoS).
Topology and protocol concerns the layout and the use of the network
resources such as routers and links. Flow control deals with effective
management of network resources for example deadlock and livelock
avoidence. QoS relates to guaranting through-put, latency or power for a
perticular transaction.
- The physical layer carries the actual electrical signals representing
the traffic. The course describes various means to overcome the challenges
of submicron technology at this layer, such as low-swing drivers,
differential signaling and asynchronous techniques.
Many options are available for networked communication. In conclusion,
the judgment of trade-off in NoC design space exploration is critical in
optimizing performance. As case studies, we evalute reasons behind the
design choices of some popular NoC architectures such as Ęthereal, Xpipes,
SPIN, etc.
Approach The module
consists of lectures and a SystemC environment in which to model
different asynchronous NoC protocols. The environment can be used for
student exercises, but no pre-designed exercises are available in the
material. Workload
The complete module amounts to a student workload of 14
hours split into
|
|