Self-calibrating asynchronous NoC links
نویسنده
چکیده
Motivation With the advent of nanoscale technologies in semiconductors, the transistor dimensions have shrunk staggeringly. Forecasts indicate that by end of year 2010, multi-billion transistors with feature size of about 50nm and clock frequency more than 10 GHz will be under design [fSI08]. As number of functional blocks in a chip increases, delays in global wires become longer than the clock period. This drives the systems-on-chip (SoC) to be decomposed into several processing cores and IP blocks. In such complex systems, the conventional bus-based interconnects fail to achieve higher bandwidth and concurrent transmissions calling for the need of communication centric design. This lead to the adoption of a new communication architecture called Networks-on-chip (NoC) [BDM02]. In NoC paradigm, each processing element is interfaced through a resource network interface (RNI) to a switch. The adjacent switches are connected to each other via bidirectional links carrying the packets of information. Later, the increasing complexity and power consumption of clock networks in NoCs lead to several researches. One of the solutions is asynchronous NoC (ANoC) where emphasis is on the division of the complex system into several blocks driven by dierent clock sources. The major challenge in ANoCs is to connect such heterogeneous blocks together and ensure robust communication. The manufacturing processes have become more complex and are dicult to control due to miniaturization which makes on-chip communication less reliable. The miniatured wires at high speed become more vulnerable to process variations. In recent days, a large number of manufactured chips are dead in the water primarily due to timing violations and causing poor yield. The asynchronous design is a way to overcome the variation eects after fabrication and to salvage chips that would otherwise be thrown away. Shrinking interconnects impose limitation on operational margins of NoCs. This lays more responsibility on design techniques to exhibit robustness to the foreseeable variations in manufacturing of NoCs. Currently used traditional design practices are based on worst-case approach. Such conservative estimates limit the speed of operation in ANoCs and 1 increase energy consumption. That is, worst-case design leaves the component to operate at region well below its maximum capability in which case the benets of the technology generation are not fully utilized. As a solution, this report details Self-Calibrating Design which allows link to autonomously calibrate against process variations. Figure 1 illustrates the relationship between supply voltage and delay under three dierent perspectives. The curve Nominal …
منابع مشابه
A Generic and Extensible Spidergon NoC
The Globally Asynchronous Locally Synchronous Network on Chip (GALS NoC) is the most efficient solution that provides low latency transfers and power efficient System on Chip (SoC) interconnect. This study presents a GALS and generic NoC architecture based on a configurable router. This router integrates a sophisticated dynamic arbiter, the wormhole routing technique and can be configured in a ...
متن کاملOn-Chip Measurement and Compensation of Timing Imbalances in High-Speed Serial NoC Links
This paper presents techniques for measurement and compensation of timing variations in clock and data channels of source-synchronous high-speed serial network-on-chip (NoC) links. Timing mismatch measurements are performed by means of asynchronous sub-sampling. This allows the use of low quality sampling clocks to reduce test hardware overhead for integration into complex MPSoCs (Multiprocesso...
متن کاملFast Asynchronous Bit-Serial Interconnects for Network-on-Chip
The multiple wires required for on-chip bit-parallel interconnect in large systems on chip (SoC) occupy large chip area and present a significant capacitive load. The problem is exacerbated in Networks-on-Chip (NoC), which employ numerous multibit links with widely varying throughput demands, activity rates and standby periods. We approach this challenge with high speed on-chip serial interconn...
متن کاملLink Pipelining for an Energy-Efficient Asynchronous Network-on-Chip
Wire latency across the links of a NoC potentially limit available bandwidth, especially in deep submicron technology. Pipeline buffers may be placed on long links to increase throughput and flit buffering. In asynchronous (clockless) NoCs, this can be done to only those links that will yield benefits, and is especially useful in heterogeneous embedded SoCs with irregular layouts and traffic. W...
متن کاملEnergy-Efficient Design of an Asynchronous Network-on-Chip
Portable electronic devices will be limited to available energy of existing battery chemistries for the foreseeable future. However, system-on-chips (SoCs) used in these devices are under a demand to offer more functionality and increased battery life. A difficult problem in SoC design is providing energy-efficient communication between its components while maintaining the required performance....
متن کامل