1. Field of the Invention
The present invention relates to an apparatus for rapid model calculation for pattern-dependent variations in a routing system. In particular, this invention relates to an apparatus for rapid model calculation having learning or self-adapting mechanisms.
2. Description of the Related Art
Routing pattern can greatly impact an integrated circuit IC in many ways. The major impact is “variations” that can degrade both yield and timing of a design. Such variations can be systematic or random. Some use variation to refer to system behavior and variability to refer to random behavior. In this invention, we do not differentiate them. We use the terms interchangeably herein. When calculating for selected model, which describe an important aspect of IC design, such as yield or timing, different routing patterns often results in different outputs. Such difference is referred to as variation hereafter.
The process window of a given routing can be affected by its neighbors. Defined by a Bossung Plot with attributes such as critical dimension (CD), depth of focus (DOF), exposure latitude (EL), the process window of a routing pattern represents how “easy” a given routing can be manufactured properly.
The timing values of a wire can be affected by its neighbors due mainly to process variations, RET steps and cross coupling.
A common way for model calculation is by a so called table look-up technique, where a large set of pre-computed input and output are stored. The calculation of the table data can be done separately, and often done offline. The real-time model calculation thus becomes simple table look-up. No learning or adapting mechanism is involved in the calculation. There is no a good outputs from the existed model to control variations from manufacturing process.
One particular aspect of the present invention is to provide an apparatus for rapid model calculation for pattern-dependent variations in a routing system. The invention details apparatus for rapid calculation of models for routing patterns. The calculation comprises learning or self-adapting mechanisms to gradually improve its accuracy. The outputs from such calculation can be used by a routing system to select routing patterns to control variations from manufacturing process. Depending on the model selection, the application areas include but not limited to yield, process window, and timing variations.
A further particular aspect of the present invention is to depict an apparatus to perform rapid model calculation that can be used in a router system. The calculator includes a learning or adapting mechanism to improve its accuracy as it continuously evolves. A knowledge base is often used to keep track of the knowledge learned.
In a preferred embodiment, the apparatus including a model calculation system comprises at least one input including a routing pattern of an integrated circuit layout, at least one output, and a calculator having a learning mechanism for continuously improving accuracy. The calculator receives the input and process the input by using the leaning mechanism to generate the out.
For further understanding of the invention, reference is made to the following detailed description illustrating the embodiments and examples of the invention. The description is only for illustrating the invention and is not intended to be considered limiting of the scope of the claim.
The drawings included herein provide a further understanding of the invention. A brief introduction of the drawings is as follows:
The invention depicts an apparatus to perform rapid model calculation that can be used in a router system. The calculator includes a learning or adapting mechanism to improve its accuracy as it continuously evolves. A knowledge base is often used to keep track of the knowledge learned.
Two types of calculators are used in the model calculation system, namely accurate calculator and rapid calculator. The former can be slow while the latter can be less accurate. The system consists of two operation modes: a learning mode and an application mode. In the learning mode, the accurate calculator produces outputs served as a reference that are fed into the rapid calculator along with its output. The rapid calculator may produce outputs that are inaccurate initially. A knowledge base is gradually built up after many learning iterations. The accuracy improved as the knowledge grows. When the rapid calculator can consistently produce fairly accurate output, it then enters the application mode. In the application mode, while the learning process can still continue with self feedback, the accurate calculator is no longer needed. The rapid calculator can produce fairly accurate output rapidly based on its continuously evolving knowledge base.
The knowledge base is both technology and model dependent. A knowledge base must be built for each process technology and for each model selected, and is obtained offline in advance. To do calculation for a new technology or a new model, one just needs to replace the underlying knowledge base.
The rapid model calculator 430 consists of three layers. The input layer 431 has two types of inputs, namely, the routing pattern 410 and the output of a feedback function 440. One example of the feedback function 440 is simply the difference between the two inputs namely, accurate output 470 and output 480. The hidden or learning layer 432 can learn from past calculations and store the knowledge in its knowledge base 460. The knowledge base 460 is technology dependent. For each process technology and each model, a knowledge base is required. The output layer 433 can produce output rapidly.
One way to speed up the learning process is to use parallel processing concept. To speed up the learning process, it is possible to divide the set of inputs into smaller subsets. Each subset of inputs can execute in parallel using the method described in
Furthermore, the model calculator also uses a curve-fitting algorithm and/or interpolation technique to speed up the learning process.
The model calculator described in this invention can be implemented in different ways. We use a Neural Network shown in
1 if W0*10+W1*11+Wb>0
0 if W0*10+W1*11+Wb<=0
In learning mode, the Network learns according to current input and the difference between output and reference output. In the application mode, the Network can produce fairly accurate output rapidly.
Therefore, the apparatus for rapid model calculation for pattern-dependent variations in a routing system uses a learning or self-adapting mechanisms to gradually improve its accuracy. The outputs from such calculation can be used by a routing system to select routing patterns to control variations from manufacturing process.
The description above only illustrates specific embodiments and examples of the invention. The invention should therefore cover various modifications and variations made to the herein-described structure and operations of the invention, provided they fall within the scope of the invention as defined in the following appended claims.
This patent application claims the benefit of the earlier-filed U.S. Provisional Patent Application entitled “Methods and Apparatus for Rapid Model Calculation for Pattern-Dependent Variations in a Routing System”, having Ser. No. 60/748,447, and filed Dec. 7, 2005.
Number | Date | Country | |
---|---|---|---|
60748447 | Dec 2005 | US |