Networks on Chip A Paradigm Change ?

1
Networks on ChipA Paradigm Change ?

• Axel Jantsch
• Laboratory of Electronics and Computer Systems,
  Royal Institute of Technology, Stockholm

2
Challange

• How do we organize a chip with one or several
  billion of transistors?

3
Why cant we just continue?Trends and Challenges

• Deep submicron effects
• Communication versus Computation
• Global synchrony
• Design productivity gap
• Heterogeneity of functions

4
Arbitrary Composability

• Given Set of components C and combinators O
• A1, A2 assamblages of components
• B is a component
• (C,O) is arbitrarily compasable ifA1 B gt
  A2can be done with effort independent of size
  (A1)

A1
A2

gt

B
5
Network on Chip
6
What is a NoC based Platform?

• A Platform is defined by
• (A) the communication infra-structure from the
  physical level to the application level
• (B) the provided services (e.g. load balancing,
  power management, fault-tolerance, resource
  allocation, task scheduling, external I/O, etc.)
  
• (C) the design methodology mapping and
  implementing applications to the platform.

7
How does a NoC Platform help?

• Reuse
• Predictability

8
Reuse

• Components and resources
• Communication infra-structure
• Reuse of the physical design
• Lower level communication services
• Higher level communication services
• Network services (operating system and
  middleware)
• Application parts and features
• Design, simulation and prototype environment
• Verification effort

9
Predictability

• Communication performance
• Electrical properties
• Design and verification time

10
The price we pay

• Less optimal products

11
Development of NOC based systems
High-perforrmance communication systems
High-capacity communication systems
Baseband platform
Personal assistant
Database platform
Data collection systems
BACKBONE
Multimedia platform
Entertainment devices
PLATFORMS
Virtual reality games
SYSTEMS
12
A NoC Platform Based Design Process

• Configuring the platform
• Selecting resources
• Reuse of features
• Evaluation and integration

13
Conclusion

• Arbitrary composability at physical,
  architectural, and application level is necessary
• If NoC can deliver that it will significantly
  change design and architecture of SoCs

14
Status of Research at KTH

• Definition of the protocol stack up to the
  transport layer
• Design of a buffer-less switch and a hot-potato
  routing technique
• NoC simulator
• Fault model and error handling scheme
• Development of a quasy-sycnhronous clocking
  scheme
• Development of a NoC Assamble Language