A design library is a collection of cells representing logic structures and certain parametric structures that designers use to implement the functions of their ASIC. The economic and efficient accomplishment of ASIC design work depends heavily upon the choice of a design library. To help designers meet design goals, we advise the evaluation team to analyze the match between the vendor's technology and the design requirements.
Performing a thorough analysis will ensure that the library contains all the necessary elements for the required ASIC function and performance. We recommend that the team consider the following factors when evaluating a design library.
There are two types of macrocells, hard and soft. Soft macrocells are functions comprised of primitive cells, which are placed and routed along with the rest of the chip. No cell layouts exist for the soft macrocells. Designers can configure soft macrocells at the time of instantiation. For example, an N-bit binary counter soft macrocell may include the following options:
Hard macrocells implement functions using custom design, usually to achieve better performance and transistor densities. The vendor tests and verifies both the hard macrocell layout and its function. Standard cells usually use hard macrocells but in some special cases gate arrays may also use them. A hard macrocell provides speed improvement over a functionally equivalent soft macrocell. Thus the hard macrocell occupies less area.
When evaluating a vendor, the function and performance of a vendor's cell library play a very important role. Macrocells are big time savers. If a vendor has a desired macrocell available, that can mean a more efficient design and considerable time saving. Designers should list the desired functions they plan to use in a design. The evaluation team can then compare this list to the functions offered by a vendor's design library.
Case Study: In selecting an ASIC vendor for the Cassini project AACS subsystem, JPL sought a vendor with a macrocell for Manchester encoder/decoder, which was a functional requirement for the ASIC. Managers knew that fulfilling encoder/decoder requirement with a macro would save considerable design time.
To economically support the test methodology of a design such as scan-based design, the cell library must have the appropriate cells. It is usually possible to implement a test approach using primitive cells, however, the additional time and gates required make this an uneconomical approach. Therefore, the evaluation team should make sure they evaluate the vendor's cell library for test features.
Although a vendor may have a large number of macrocells, some of the cells may not support the vendor's chosen testability approach. The evaluation team should make certain that all the cells in a design library support the vendor's methodology for testing ASICs.
If a vendor supports a scan design methodology for design for test (DFT), then make sure the design library has the following features available:
The vendor may support a tool for automatically generating stuck-at fault test vectors, also called "automatic test pattern generation" (ATPG). If available, the evaluation team should review this capability, for:
For radiation tolerant design, the evaluating team needs to verify that the vendor offers the following features in his design library:
MIL-STD-883 mandates that manufacturers have a test chip to verify performance characteristics of the library cells. To conform with this standard, the vendor must have verified all setup, hold, minimum cycle time, enable and disable times provided with simulation models, through either circuit simulation or parameter extraction from actual hardware tests. The library analysis must include verification of the software-based library against the structures as actually built.
We suggest the evaluating team examine whether a macrocell has been implemented in silicon and verified for functionality and performance limits via actual hardware tests or whether the vendor's policy calls for only verifying the macrocell through software simulation.
The team also needs to determine what model levels the vendor has available for test and verification (characterization) of library cells. Examples of these model levels are:
The levels used for evaluation tell the evaluation team how thoroughly the vendor checks new cells before bringing them into his libraries and the chances of a problem developing because of a difference between cell models and the cell as built. Ideally, all levels are examined for correct function and parametric performance before a cell joins a library.
The team should examine evidence showing the vendor has verified all the specifications in the library data book. The team should also examine the vendor's actual use of regular regression testing to ensure new problems have not occurred and old problems have not crept back. All the tests and test results must be available upon request.
The cell library data book, at minimum, should provide the following information:
Some vendors provide a user's manual for the library. This manual helps designers by furnishing techniques for designing reliable ASICs and designing for performance and density.
Now you may jump to: