In Analog/RF layouts, designers frequently use slotted metal structures. Such slotting is done either to satisfy DRC requirements from foundries to satisfy max. density rule criteria, or to reduce eddy current losses in return paths of a transmission line/coiled spiral inductor. Current flow in such slotted metal structures is non-uniform, hence, traditional parasitic extraction approaches with predefined fracture length specifications do not result in accurate pin-to-pin resistance. Hence, Cadence Design Systems sign-off extraction solution – Quantus, provides a mesh extraction approach for these cases, where the slotted/wide metal structures are broken down into smaller squares each representing a small parasitic resistance. This approach of moving from a lumped to distributed model for resistance extraction leads to much better DC V/I current modelling. For example, the MIMCAP layout structure shown below has its 2 terminals ctm (top) and cbm (bot), connected to Metal 13 for its VSS and VDD connections. Its ESR (Effective Series Resistance) is calculated at 50GHz by doing an AC V/I simulation.
Image Source: PVS Quick View Editor
ESR @ 50GHz with regular square counting based extraction is 19.04 ohms (incorrect), whereas by using advanced adaptive mesh gives 5.16 ohms (correct). The MIMCAP structure has very wide top and bottom terminals which needs to be modeled using mesh approach to give accurate results. However, the challenge with this approach is increased netlist size, simulation run time and user overhead in setting up the mesh layers and corresponding mesh sizes to be extracted.
Quantus has newly introduced advanced adaptive mesh extraction, where intelligent meshing is applied to automatically detect non-uniform current regions near discontinuous/slotted metal structures where adaptive grid meshing is applied and detect uniform current regions near continuous metal structures where square counting-based meshing is applied.
The below case study is performed on a ST BiCMOS 55nm process. However, the advanced adaptive mesh technology can be applied to other advanced technology nodes too with similar benefits.
Image Source: Cadence Virtuoso Layout Editor
Adaptive mesh fractured the M1 slotted ground plane by using smaller fracture regions even in right most regions of the layout structure, where there are no slots and current flow is uniform. Whereas, the new Advanced adaptive mesh (available in latest Quantus builds) was able to automatically distinguish between the slotted and continuous metal regions and apply intelligent meshing only near the slotted metal portion, where current flow would be non-uniform and distributed modelling is required. Also, in Advanced adaptive mesh, the size per mesh is 5.96x5.96 which is larger than the 2.59x2.59 determined automatically in Adaptive mesh. Since each fractured square region corresponds to 1 parasitic resistor, this helps reduce the total number of parasitic resistors. A combination of these 2 techniques, helped reduce the netlist size dramatically by ~34%.
In cases where only a certain region of the layout structure needs distributed modelling, meshR user region can be defined, so that Quantus performs advanced adaptive or adaptive meshing in only a certain portion of the layout, whereas the remaining would be square counting based fracturing approach. This further helps reduce netlist size.
The reduction in number of parasitic R and C’s in the advanced adaptive mesh netlist as a result can speed up downstream simulation run time while consuming lesser memory. Its critical to also ascertain the accuracy of the reduced netlist wrt. adaptive mesh to confirm the reduction doesn’t come at the cost of accuracy. Hence, a pin-to-pin DC resistance check is performed on few fractured sub-nodes from the 2 DSPF netlists, keeping in mind the sub-nodes are picked with roughly similar x,y coordinates. Advanced adaptive mesh can also be applied to spice extracted view output formats. As demonstrated, pin to pin resistance accuracy in advanced adaptive mesh is highly accurate.
Depending on the level of simulation accuracy needed, there is a capability to specify a k-scaling factor to increase or decrease the auto mesh size. This helps fine tune the netlist size vs. simulation accuracy desired.
The Adaptive Mesh feature is based on the user-specified mesh size for the conductor layers that are to be meshed. Alternatively, Advanced Adaptive Mesh feature is based on automatic mesh sizing for the conductor layers. Thus, setup and usage of Advanced adaptive mesh is simpler for these users. With the advent of advanced nodes and slotted layout structures, distributed MIMCAPs designs and other newer design artifacts, it is recommended to use the Advanced adaptive mesh solution in Quantus to achieve better netlist size, accuracy and simulation run times.