A Methodology for Constraint-Driven Synthesis of On-Chip Communications
read more
Citations
A power-aware mapping approach to map IP cores onto NoCs under bandwidth and latency constraints
Design of an Energy-Efficient Asynchronous NoC and Its Optimization Tools for Heterogeneous SoCs
The Future of Formal Methods and GALS Design
Comparing Energy and Latency of Asynchronous and Synchronous NoCs for Embedded SoCs
CusNoC: Fast Full-Chip Custom NoC Generation
References
Networks on chips: a new SoC paradigm
Route packets, not wires: on-chip interconnection networks
Deadlock-Free Message Routing in Multiprocessor Interconnection Networks
Deadlock-free message routing in multiprocessor interconnection networks
Related Papers (5)
Frequently Asked Questions (12)
Q2. What are the two operations that allow the design of complex on-chip communications?
To allow the incremental design of complex on-chip communications, the authors introduce two operations: renaming and parallel composition.
Q3. What is the restriction of a configuration to a subset of components?
Given a communication structure N(C,q, L), the restriction of a configuration l ∈ L to a subset of components C′ ⊆ C, denoted by l|C′ , is a function f : C′ → Dq such that f(c) = l(c) for all c ∈ C′.
Q4. How is the power of the noC found by the heuristic?
The power of the NoC found by the heuristic is within 2x from the power found by CPLEX that is very optimistic for the change in the cost function and for the relaxation of the integer constraints.
Q5. How many bits of flit were used to achieve a maximum link capacity of 3.2?
Since the total memory bandwidth is 3GBps, the authors set the flit width to 128 bits to achieve a link capacity of 3.2GBps with a maximum flit rate of 200 ·106 per input port of the routers.
Q6. Why is the maximum number of nodes limited to lI?
Because the authors want lI [(x, y, τ)] to be injective (i.e. only one component of a specific type can be installed in a particular location), the maximum number of nodes in any platform instance is limited to |D(x,y,τ)|.
Q7. In what section does the optimization technique explore the isomorphic-free set of regular top?
In [24], the optimization technique explores the isomorphic-free set of all regular topologies and in [25] the authors assume that one NP is given as input to their algorithm.
Q8. What is the communication structure among all possible platform instances?
According to Lemma 1, if the authors can find the greatest element NP of 〈L〉 with respect to the ordering relation ≤qP , then the solution of problem PR1 with NP = NP is the best communication structure among all possible platform instances.
Q9. How many libraries of communication components were used in this experiment?
The authors used six libraries of communication components differing for the flit-width of the data path (32 and 128 bits corresponding to 280 ·106 and 70 ·106 flits per second, respectively) and the size of the largest switch available in the library (2 × 2, 5 × 5 and 8 × 8).
Q10. What can be used to optimize the bus circuitry?
The transfer table information can be used at a lower abstraction level to optimize the bus circuitry (e.g. decoders and multiplexers) or even to segment the bus and insert bus bridges.
Q11. What is the procedure that checks the delay constraints of the nodes?
If a delay model must be taken into account to check delay constraints, the best path is discovered by a labeling algorithm (SpLabeling) that finds the minimumcost constrained shortest path between two nodes; a modified version of Dijkstra’s shortest path algorithm is used otherwise.
Q12. What is the degree constraint for the rip-up and reroute approach?
The links connected to the output of nodes with output degree violations and links connected to the input of nodes with input degree violations are the ones that are considered for rip-up and re-route.