ChipEstimate.com / Cadence - update since acquisition
May 30th, 2008 by adminI had an opportunity to meet with Adam Traidman of ChipEstimate.com, which is now part of Cadence. The discussion centered around some of the issues and trends in sub-wavelength (130nm and below) process node IP and included an update on the status of ChipEstimate.com.
Current status of the product since the acquisition: the former Chip Estimate InCyte product is now available for sale from Cadence as the Cadence InCyte Chip Estimator and Cadence Chip Planning System. It is available in two flavors, one for mainstream fabless companies and the other for IDM’s and IFM’s (Integrated Fabless Manufacturers). The portal still has a free version (InCyte Lite) with some reduced features available for design service, startups and budgetary planning applications.
Current status of the portal since the acquisition: the ChipEstimate.com Portal is being maintained as an independent offsite portal for IP providers and information for SOC estimation/power planning and there is no anticipated change to that status. Even though Cadence will not be at DAC this year, the portal will be exhibiting and offering a series of talks from foundries and hard & soft IP providers dealing with SOC integration and IP use issues. The presentations will be provided by over 20 of the 175+ IP suppliers who have offerings detailed on the ChipEstimate.com portal.
On the topic of sub-wavelength SOC build and IP Selection, we identified 6 gating common problem areas with respect to the standardized use of IP from multiple suppliers. These issues are:
(2) Design data grid and its applicability to the tool flow used. As most SOCs have multiple tool flows (custom IP level, Megacell level and chip assembly level), commonality of the design grid is required for predictable performance and proper implementation of DFM / DFY correction.
(3) Abstract description, routing and blockage description. Most IP utilizes a small number of interconnect layers (2-5 metals), and the SOC’s are on processes that support up to 12 layers of metal. Application specific blockages and pre-routes must be described in order to keep the characterization of the IP from being comprimised.
(4) Multi-mode power methodology. There are a number of application based power modes (full power, standby, hibernate, display off, etc) as well as implementation strategies to achieve these modes (Switched rails, MTCMOS, etc). It is important to identify which power modes are supported by the IP providers so blocks with a common power strategy can be grouped together and avoid hundreds of thousands of gates of power control logic at the SOC level.
(5) Memories and Mixed Signal IP have parametric constraints well above and beyond traditional timing and power issues. As a result, placement of these blocks, specification of adjacencies and cross wafer variability must be addressed in addition to the nominal operational specification.
(6) I/O libraries are typically provided by both foundries and third party IP suppliers. Construction of the I/O ring with multiple power rails, ESD, simultaneous switching characteristics, ring break points and isolation regions, corner requirements, signal level shifting and logic vs memory vs analog pads is a major challenge. Currently, there is no automation or systematic meta-data methodology for transferring this information, and it relies heavily oa a service based support model and paper documentation.
PC





