Computer-aided design and analysis of circuits and semiconductor – Nanotechnology related integrated circuit design
Reexamination Certificate
2002-08-28
2004-09-07
Siek, Vuthe (Department: 2825)
Computer-aided design and analysis of circuits and semiconductor
Nanotechnology related integrated circuit design
C716S030000, C716S030000
Reexamination Certificate
active
06789233
ABSTRACT:
TECHNICAL FIELD
The present invention relates to a method for determining a matched routing arrangement for a semiconductor device. More particularly, the invention is directed to a computational method using a genetic algorithm to determine a matched routing arrangement.
BACKGROUND OF THE INVENTION
The significant advances in computer technology and in electronics in general is due in large part to the development of microelectronic devices of increasing density and smaller scale. The smaller and more complex semiconductor circuits formed within the microelectronic device generally include discrete elements such as resistors, capacitors, diodes and transistors that are organized into separate cells to perform a specified function. Each of the cells in the microelectronic device may have one or more pins, which may be coupled to one or more pins on another cell by wires extending between the pins. In the discussion that follows, pins are understood to refer to contacts, or coupling points for the wires. The wires coupling the cells are generally formed in a layer of the device that contains the cells. A net is a set of two or more pins that must be coupled by wires. In many cases, a net includes only a pair of pins that require coupling. In some cases, however, the net may include three or more pins that must be coupled. A list of nets for the microelectronic device is commonly referred to as a netlist.
The lengths of the nets that couple the pins is a critical factor in the design of microelectronic devices, since the propagation of a signal along an interconnecting net requires a finite length of time. Generally, when the net layout is designed, a significant effort is made to minimize the lengths of the nets. As the scale of microelectronic devices continues to decrease, however, the relative delays in signal propagation incurred by nets of unequal length has become a significant factor affecting the performance of these devices, which is not adequately addressed by minimizing the lengths of the nets.
A determination of the net lengths extending between the pins to achieve either a shortest net length solution, or alternatively, to match the net lengths, may be performed in a variety of ways, including a simple hand calculation of the routing or other approximate methods that may be implemented during the net layout stage. In general, however, the determination of either the minimum net length solution, or the determination of a matched net length solution for a microelectronic device is a complex multivariable optimization problem that is best suited for solution using a computer, which can select optimum net lengths based upon repeated calculations of actual net lengths. If the number of pins becomes large, however, the foregoing “brute force” solution method becomes increasingly intractable. For example, in calculating the net lengths in a minimum net length solution that includes N different pins, testing every possible combination is estimated to require on the order of about N! additions. For a problem having only nine pins, and assuming an arithmetic speed of 1 billion additions per second, an estimated computation time on the order of approximately about 0.4 milliseconds is required. If the number of pins is doubled to 18, however, the estimated calculation time is on the order of about 0.2 years. Still further, if only a single additional pin is included, the estimated calculation time dramatically increases to a time on the order of about 3800 years. Clearly, brute force calculation methods are incapable of handling optimization problems of this kind, unless the number of pins in the microelectronic device is of modest size.
One known method for solving complex multivariable optimization problems of the foregoing type is through the use of genetic algorithms. Genetic algorithms generally represent a class of algorithms that solve optimization problems by at least partially simulating the evolutionary processes of natural selection. Accordingly, favorable outcomes of an iterative computational scheme are combined with still other favorable outcomes to accelerate convergence to an optimum solution, while the less favorable solutions are discarded. Consequently, the number of discrete arithmetic operations is generally substantially reduced through the use of a genetic algorithm. Without regard to the exact nature of the optimization problem, a genetic algorithm generally proceeds through a series of steps, as described below.
A genetic algorithm procedure generally begins with an initialization step, wherein an initial population of solutions to the optimization problem is generated. The initial population may be obtained from other computational procedures, or it may consist of a series of solutions that are initial estimated solutions, or still further, the initial population may be randomly generated. Next, an evaluation is performed by applying a problem-specific evaluation function to the initial population of solutions. The evaluation function thus determines the relative acceptability of the solution. The evaluation of the initial population permits solutions to be selected as parents of the next generation of solutions. In order to advance the computation, the more favorable solutions may be selected as parents numerous times, while the less favorable solutions are not selected at all.
Each pair of parent solutions is subsequently combined through a crossover process that produces a pair of offspring that have similarities to both parents. The crossover process is central to the technique, since it allows the concentration of traits associated with an optimal solution to be manifested in a single individual. In addition, a mutation operator may be applied to either offspring. When a mutation operator is applied, some random change generally occurs to a randomly selected characteristic, so that the offspring solution is thus derived asexually. In general, most genetic algorithms utilize a combination of mutation and crossover to derive successive populations.
The offspring population is then evaluated by applying the evaluation function to each member of the offspring population. Since the offspring are modified forms of the parent population, at least some of the offspring are evaluated more favorably than the parent population from which they were derived. Further refinement may occur by combining the offspring population with the population that the parent population originated from, in order to recursively optimize the solution. Termination of the calculation can be based upon the achievement of predetermined convergence criteria, such as the difference of successive solutions differing by at most a predetermined value, or alternatively, when the average solution for the population has not changed within a predetermined number of iterations.
In general, no method currently exists to achieve a matched net length solution using a genetic algorithm procedure. Instead, layout designs having matched net lengths are generated by hand calculation or other approximate methods, which generally consider only a single net at a time. For detailed netlists, a considerable amount of time is generally required to determine net lengths, with the resultant nets being only reasonably well matched. Accordingly, there is a need in the art for a computational method that will permit net lengths to be better matched than current methods allow. Further, there is a need in the art for a computational technique that eliminates time consuming manual methods frequently utilized to provide approximate matched net lengths.
SUMMARY OF THE INVENTION
The present invention relates to a method for determining a matched routing netlist for a semiconductor device. In particular, the invention is directed to a computational method using a genetic algorithm to determine a matched netlist. In one aspect, a topological plan for the microelectronic device is compiled and used to calculate a population of netlists. The netlists are then dynamically modified using a genetic algorithm to gene
Gore Brooklin J.
Lee Michael D.
Priest Matthew L.
Dorsey & Whitney LLP
Micro)n Technology, Inc.
Siek Vuthe
LandOfFree
Method for determining a matched routing arrangement for... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method for determining a matched routing arrangement for..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method for determining a matched routing arrangement for... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3261445