Computer-aided design and analysis of circuits and semiconductor – Nanotechnology related integrated circuit design
Reexamination Certificate
2000-08-08
2002-03-19
Smith, Matthew (Department: 2825)
Computer-aided design and analysis of circuits and semiconductor
Nanotechnology related integrated circuit design
Reexamination Certificate
active
06360356
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates generally to methods and systems used to create efficient physical implementations from high level descriptions of electronic designs and, in particular, to a software system and method that optimizes Register-Transfer-Level (RTL) descriptions with respect to performance parameters including area, timing, and power, prior to logic synthesis, floorplanning, placement and routing.
2. Description of the Background Art
Present Electronic Design Automation (EDA) systems for designing electronic systems consist of software tools running on a digital computer that assist a designer in the creation and verification of complex electronic designs. Present day state-of-the-art design technique uses a combination of logic synthesis, floorplanning, place-and-route, parasitic extraction, and timing tools in an iterative sequence to form a design process commonly known as the top-down design methodology.
The left side of
FIG. 1
illustrates a typical top-down design process. The primary entry point into the top-down design flow is a high level functional description, at behavioral-level or RTL, of an integrated circuit design expressed in a Hardware Description Language (HDL). This design is coupled with various design goals, such as the overall operating frequency of the Integrated Circuit (IC), circuit area, power consumption, and the like.
Conventional top-down methodology uses two overlapping processes, a front-end flow, and a back-end flow. Each of these flows involve multiple time consuming iterations, and the exchange of very complex information. In the front-end of the top-down methodology, the RTL model is manually partitioned by the designer into various functional blocks the designer thinks best represents the functional and architectural aspects of the design. Then, logic synthesis tools convert the functional description into a detailed gate-level network (netlist) and create timing constraints based on a statistical wire-load estimation model and a pre-characterized cell library for the process technology that will be used to physically implement the integrated circuit.
The gate-level netlist and timing constraints are then provided to the back-end flow to create a floorplan, and then to optimize the logic. The circuit is then placed and routed by the place-and-route tool to create a physical layout. After place-and-route, parasitic extraction and timing tools (typically by the circuit fabricator) feed timing data back to the logic synthesis process so that a designer can iterate on the design until the design goals are met.
While the synthesis and place-and-route automation represent a significant productivity improvement over an otherwise tedious and error-prone manual design process, the top-down design methodology has failed to produce efficient physical implementations of many circuit designs that take fall advantage of the capability of advanced IC manufacturing processes. This is evident in the growing “design gap” between what semiconductor vendors can manufacture with today's deep sub-micron processes and what IC designers can create using top-down EDA design tools. The latest 0.18 &mgr;m CMOS process can fabricate silicon die with 10 million gates, running at speeds in excess of 500 MHz. In contrast, designers using conventional top-down EDA tools struggle with the creation, analysis, and verification of integrated circuits having 0.5-1 million gates, running at 150 MHz.
The primary inefficiency of the top-down methodology arises from its reliance on statistical wire-load models proved to be inadequate in wire-delay dominated deep sub-micron digital systems. Timing in deep sub-micron integrated circuits is dominated by interconnect delays rather than gate delays. Conventional top-down design tools, such as behavioral and logic synthesis, were originally designed in an era when gate delays dominated chip timing. These tools use inaccurate, statistical wire-load estimates to model wiring parasitics at early stages in the design cycle, and the effects of these inaccuracies are propagated throughout the rest of the design methodology. To overcome the timing model inaccuracies, the designer engages in excessive and time-consuming iterations of logic synthesis, floorplanning, logic optimization, and place-and-route in attempting to converge on the timing constraints for the circuit. This iterative loop is referred to as the timing-convergence problem.
The large discrepancy between statistical wire-load model and actual wire-load means that circuit designers must wait until gate-level floorplanning and place and route tasks are complete to begin chip-level optimization. The enormous gate-level complexity of today's system-on-a-chip designs places a heavy burden on gate-level verification and analysis tools and makes multiple design iterations very time consuming.
Additionally, the complexity of present high performance integrated circuit designs overwhelms the capability of logic synthesis tools. Synthesis execution times of many hours on present day high-performance engineering workstations are typical for circuits containing only tens-of-thousands of logic gates. Place-and-route execution times for these circuits can also consume many hours. It is not unusual for a single synthesis and place-and-route iteration for a circuit containing tens-of-thousands of logic gates to take days. Synthesis and place-and-route tool run times grow non-linearly, sometimes exponentially, as the size of the circuit grows and as circuit-performance goals are increased. Thus, logic synthesis cannot process complex designs all at once. Designers are forced to develop functional descriptions and manually partition the design into smaller modules, upon which logic synthesis is individually performed. During manual partitioning, however, the designer has little or no accurate information on the back-end physical effect of the partitioning, and in particular, on the effect of such partitions on timing, area, and power consumption. The relationship between high-level functional description and the low-level layout physical effect is not obvious at the front-end design stage. The failure to predict accurate back-end physical effect at or above the RTL design stage results in local optimization and a sub-optimal functional description of the design. Design efficiency suffers due to design over-constraint (timing non-convergence) or under-constraint (loss of performance and density), or some combination of both for various different partitions of the integrated circuit. Sub-optimal RTL descriptions and partitioning serve as a poor starting point for logic synthesis, which propagates and amplifies the design deficiencies, eventually leading to silicon inefficiency (e.g., excessive area or power consumption, slower operating frequency), even after long iteration and manual intervention.
Further inefficiency in the top-down design methodology is introduced because logic synthesis tools treat all logic as random logic. Consequently, logic synthesis typically fails to recognize and take advantage of more efficient silicon structures such as datapaths, which are commonly used and expressed in the high level description of the design. Designers who recognize this limitation frequently bypass synthesis by manually instantiating gate-level elements in their RTL source. This is equivalent to writing a gate-level netlist, an onerous, low-productivity, and error-prone task.
Another deficiency of the top-down methodology is that it requires a cumbersome netlist hand-off between front-end and back-end design cycles. Complex bi-directional information transfer occurs at the overlap between front-end and back-end iteration loops. The diverse design expertise required to effectively manage the top-down design process is rare and not commonly available to a typical design team. Design inefficiency causes the costly under-utilization of advanced IC manufacturing processes. The iterative nature of the top-down design methodology
Do Thuan
Fenwick & West LLP
Smith Matthew
Tera Systems, Inc.
LandOfFree
Creating optimized physical implementations from high-level... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Creating optimized physical implementations from high-level..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Creating optimized physical implementations from high-level... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2832877