Raphael NXT Full-Chip Capacitance Extractor

Overview
Raphael NXT is a true three-dimensional (3D) capacitance extractor that provides silicon-accurate self and coupling capacitances for IC design. Equipped with an ultrafast extraction engine, Raphael NXT complements Star-RCXT by extracting 3D capacitances of critical nets, cells, or blocks on the full-chip level. Raphael NXT supports the latest processes that include conformal dielectric layers, trapezoidal conductors, lithography effects, and metal fill, providing an accurate representation of the complex geometries found in interconnect structures. As a result, capacitance values extracted by Raphael NXT are closely correlated with Raphael, the gold-standard reference field solver, or silicon measurements.

Benefits

• Extraction of field solver–accurate capacitance for critical nets, cells, and blocks on the full-chip level
• Accounts for detailed process effects predominant in leadingedge nanoscale technologies
• Seamless integration with market-leading Star-RCXT full-chip parasitic extractor
• Flexibility to run Raphael NXT on a single CPU or in parallel on many machines across a network

Description
The performance of circuits fabricated at the 130 nm node and below is dominated by the parasitic capacitance and resistance of interconnects. Therefore, accurate parasitic extraction is critical to first-pass silicon success. While mesh-based field solvers are very effective at extracting small structures such as SRAM cells, larger nets require a more efficient solver.

Raphael NXT solves the Laplace equation using the floating random-walk method. This method addresses problems of size well beyond the reach of mesh-based field solvers, while accounting for all 3D effects as defined by the geometric information entered into the tool without compromising its accuracy.

Detailed process effects – conformal dielectrics, trapezoidal conductor cross sections, and lithography effects – can be handled, and the tool can extract floating metal (metal fill) without introducing extra nodes in the SPICE netlist.

Figure 1: Interconnect extraction hierarchy. Raphael NXT combines the accuracy of mesh-based field solvers with an efficient algorithm capable of extracting hundreds of nets.

Raphael NXT provides highly accurate capacitance extraction of critical nets, cells, and blocks, complementing Star-RCXT on the full-chip level as part of an overall interconnect extraction tool hierarchy (see Figure 1).

The statistical nature of the algorithm means that results are reported with statistical confidence limits. As the number of sampling random walks increases, the reported solution approaches the silicon value.

Therefore, there is a trade-off between the statistical uncertainty of the solution and CPU time. The user can choose the default tolerance of 1 s at 3% total capacitance (the error in the total capacitance for 68% of the extracted nets is less than 3%) or can specify the desired accuracy for self and coupling capacitance.

Raphael NXT results have been correlated with the gold-standard capacitance extractor Raphael using a set of structures that can be addressed by both solvers. The correlation yielded a mean error of less than 0.3% and a standard deviation in the error of 1.3%.

Using Raphael NXT with Star-RCXT
Star-RCXT, the industry’s leading full-chip parasitic extraction tool, uses pattern-matching algorithms based on interconnect structures calibrated with field solvers. In cases when the highest accuracy is required, as in clock trees and other critical nets, Star-RCXT interfaces seamlessly with Raphael NXT in two modes: FSCOMPARE and FS_EXTRACT_NETS.

The FSCOMPARE command provides an automated push-button flow for the analysis of test structures or selected nets from a chip using Raphael NXT. Then, the results from Raphael NXT and Star-RCXT are compared to ensure consistency between the extractors.

Within the Star-RCXT environment, users can utilize the FS_EXTRACT_NETS command to supply Raphael NXT with a list of nets that need the highest level of accuracy for capacitance extraction. Star-RCXT not only extracts the nets as in the regular flow, but also creates a subset of the design (the fs.cap file) based on the nets listed in the FS_EXTRACT_NETS command to be extracted also by Raphael NXT. The resulting netlist is a mixture of Star-RCXT extracted nets and Raphael NXT extracted nets (see Figure ).

Figure 2: Interconnect extraction flow. The full-chip extraction tool Star-RCXT uses a database of 2D field-solver solutions for its pattern-matching engine. During extraction, critical nets are analyzed with Raphael NXT and are compared to pattern-matching extraction to verify accuracy.

Advanced Process Modeling
When high accuracy is needed, detailed process effects must be included in the interconnect structures analyzed by Raphael NXT. Dummy metal fill used to even out pattern density in chemicalmechanical polishing of copper conductors can be extracted as floating metal without introducing extra nodes in the output SPICE netlist.

In dual-damascene copper processes, the cross section of the conductors is more trapezoidal than rectangular. For the small pitches used at 90 nm and below, this effect can be significant. In addition, low-k dielectric films are often integrated with higher dielectric constant etch stop and conformal layers. Raphael NXT supports interconnect file formats that allow users to describe these complex interconnect stacks as shown in Figure 3.

Figure 3: Raphael NXT supports advanced process features such as (a) conductors with trapezoidal cross sections and (b) conformal dielectric layers.

The extraction engine in Raphael NXT is not confined to rectangular Manhattan geometries. In advanced applications, output from a lithography simulator can be combined with a corresponding technology file to create interconnect structures for Raphael NXT that incorporate optical proximity effects.

Boundary Conditions
By default, Raphael NXT uses a Dirichlet boundary condition of 0 V (electrical ground) at the edges of the simulation domain. However, users can specify Neumann (reflecting) boundary conditions explicitly. The tool also supports periodic boundary conditions to handle geometries with repeated cells such as RAM devices and charge-coupled devices.

Figure 4: Capacitance correlation between Raphael NXT and Raphael for 513 test structures for a 65 nm process. The correlation coefficient is 0.99787.

Figure 5: Scaling of solution time using distributed processing capability with 1, 2, 4, 6, and 8 processors taken from a 90 nm design in which 100 nets were extracted.

Correlation with Raphael
Raphael NXT results have been correlated with the capacitance solver Raphael. Figure 4 shows a correlation plot for over 500 test structures. Though these tools address different applications, this correlation ensures consistency of results for solvers using different solution algorithms.

Multi-CPU Processing
One of the key advantages of the floating random-walk method is its inherently parallel characteristics. Raphael NXT includes a distributed processing option to provide users with a scalable high-performance interconnect extraction platform (see Figure 5). The implementation uses a fine-grained parallel algorithm, which runs efficiently even with different loads on different machines. As designs become larger and more nets require the accuracy of Raphael NXT, users can easily add more CPUs to keep the total extraction time manageable.

Inputs

• Design file (GDSII)
• Technology file
• Control file (optional)

Outputs

• Capacitance report
• Capacitance matrix
• SPICE netlist (SPF format)

Supported Platforms

• AMD64 64-bit RHEL v3
• Sun 3 -bit Solaris 9
• Sun 64-bit Solaris 9
• x86 3 -bit RHEL v3

For more information about Raphael NXT and other Synopsys TCAD products and services, go to www.synopsys.com/products/tcad/tcad.html, or contact your local Synopsys representative, or email tcad_team@synopsys.com.