SolutionWare: Characterization

"From using your test-bench, we found that our behavioral model had a small problem. But, your test-bench also lead us to finding a real problem with our design related to power-down modes. This basically saved us from having to do a re-spin on our ASIC!!" Jon Rose, Innovasic Semiconductor
Optimized for Performance
Solutionware library generation tools offer the most comprehensive set of tools for characterization and model generation. They are optimized for speed and efficiency. They can cover even non-standard cells like one-hot muxes, semi-synchronous flops, synchronizers, clock gating cells, dynamic flops and dual edge flops, and even cells which mux between flip-flop and latch behavior. Characterization time for a flip-flop could be as small as 2 minutes for 7x7 tables generated without interpolation.

\bgroup\color{blue}$\includegraphics{} $\egroup

Functional Testing
SpiceTest is a Spice level functional verification tool. It takes as input a spice netlist implementation of the cell and a functional specification in the form of ACDL, generates test vectors and compares the outputs from spice simulation against the expected values. SpiceTest is usually invoked automatically by MakeLib. In one sense, it carries out all the functional tests which should be carried out by the circuit designers to verify the functionality of the cell. SpiceTest should be used before characterization to verify that the circuit has been designed per functional specification, and that no translators or circuit extraction tool have altered its functionality. It can also be used during the design stage of the cell, where it relieves the user from the laborious task of SPICE deck preparation. The vectors generated by SpiceTest provide 100% functional coverage. In addition to verifying that the cell is functioning correctly, SpiceTest also checks if the characterization environment for this particular cell is set up correctly. If the cell is going to be overloaded, or is expected to switch faster than it is capable of doing, this tool will report it.

The cell is embedded in a test circuit where it is driven by ``real'' live drivers and loaded with the maximum loads at its outputs. Automatically created SPICE stimuli are applied and the outputs of the circuit are compared against the expected values. Especially for new cell libraries, SpiceTest has been very effective in identifying faulty cells.

ACDL description of the cell is the specification upon which inputs and expected values of the outputs are based. Even for complex cells, test vectors can be generated in seconds. The Cell-Under-Test is driven by buffers after loading its inputs and outputs. This setup tries to be as close to In-Circuit-Testing as possible. SpiceTest generates a SPICE netlist which includes the model cards, environmental parameters, test vectors generated from ACDL, an instance of the Cell-Under-Test, its drivers and loads. Finally, SPICE is called to do circuit level simulation. Its output is compared to the expected values at strobe points determined by SpiceTest. Any unstable outputs and functional errors are reported to the user. A SpiceTest user does not really need to know how to use SPICE and do circuit simulation. The cell design and verification process can be accelerated by delegating the verification part to SpiceTest.

Before characterization starts, various test vectors are analyzed for state-dependency. State-dependent delay, power, input capacitance as well as setup/hold conditions are identified by StateGen.

Automatic Stimuli Generation
One of the unique features of Solutionware is its integrated test vector and stimuli generator. User does not need to specify any stimuli to be passed to circuit simulator. It is all automated. The functionality of the basic cells is quite simple, and there are few inputs. However, when the cell is implemented at the circuit level, it is quite possible that it may not be implemented correctly. This is especially true for complex flip-flops, though we have seen quite a number of combinational cells not matching their expected behavior. By separating the description and circuit level implementation, an extra layer of reliability checking could be added to the cell design process. The most time consuming and error prone activity in the library creation process is the stimulus generation. StimGen eliminates this bottle neck by synthesizing the stimuli from a higher level description of the cell, ACDL, which is also the entry point for most of the LTI tools. StimGen generates inputs for LibChar, the characterizer. These are generic spice inputs, which are independent of the circuit simulator, of the supply voltages and various timing parameters. For almost all ASIC library cells, including I/O and scan cells, the stimuli can be generated automatically. Stimuli for measuring the following parameters can be generated automatically for any cell which can be described in ACDL, which partially include
\bgroup\color{blue}$\bullet$\egroup Input to Output Path Delays \bgroup\color{blue}$\bullet$\egroup Output to Output Path Delays
\bgroup\color{blue}$\bullet$\egroup Three-State Enable/Disable Path Delays \bgroup\color{blue}$\bullet$\egroup Setup/Hold Times
\bgroup\color{blue}$\bullet$\egroup Recovery/Removal/Nochange Times \bgroup\color{blue}$\bullet$\egroup Nochange Times
\bgroup\color{blue}$\bullet$\egroup Clock/Enable Pulse Width \bgroup\color{blue}$\bullet$\egroup Maximum Skew
\bgroup\color{blue}$\bullet$\egroup Input Pin Capacitances \bgroup\color{blue}$\bullet$\egroup Input Pin Hysteresis
\bgroup\color{blue}$\bullet$\egroup Input Toggle Power \bgroup\color{blue}$\bullet$\egroup Input/Output Propagation Power
\bgroup\color{blue}$\bullet$\egroup Output Current Drive \bgroup\color{blue}$\bullet$\egroup Static Power/State dependent

From analysis of an ACDL description of the cell, StimGen determines what the appropriate parameters are and writes out stimuli accordingly. Any cell with at most one internal state, irrespective of the number of inputs, and its functional complexity can be handled automatically. For multiple internal states, multiple equations are required in the ACDL description.

Characterization Engine: LibChar

Inputs to LibChar are the SPICE netlist of the cell, which can be extracted from layout, and a stimulus file describing each parameter to be calculated. Stimuli are generated by StimGen automatically from the ACDL \bgroup\color{blue}$^{\mbox{\tiny TM}}$\egroup description of the cell and the state-dependency information from StateGen. The actual values of the supply voltages, temperatures, SPICE models to use, delay models, threshold voltages are specified in LibChar configuration files, .libcharrc. All these variables can be grouped into various corners.

\bgroup\color{blue}$\includegraphics{} $\egroup
The cell to be characterized is embedded into a test structure where it is driven by the real live drivers. This helps to capture the coupling between a cell and its drivers on silicon properly, and also eliminates any unrealistic assumptions about the shape of the input waveforms. The inputs buffers are loaded with fixed capacitors to generate different input slope conditions. Outputs of the cell are loaded with different capacitive loads while measuring input to output delays. In addition, the user can specify different types of input drivers.

There are several ways of measuring pin-to-pin delays and transition times at the user's disposable. Same setup is used for measuring power, supply currents and input loads. LibChar can optionally produce CCS, ECSM, CCS-power and signal integrity data. Execution time is highly optimized. Various algorithmic techniques are employed to reduce the number and duration of SPICE runs. LibChar and its friends provide the most robust and reliable characterization and modeling environment. Same set of tools are used across cell, IO and memories, providing a consistent set of features and capabilities across all of them. ACDL is especially suitable for handling non-conventional cells, and good for any type of logic family including dynamic and current mode logic families.

Most characterization solutions provide a fixed set of parameters to be associated with each cell. Typically these are generated manually as well, it becomes difficult to verify they are what they are claimed to be. LTI tools maintain one internal database of all timing and power parameters, and all models are generated from this database. It is compared with the behavior of the circuit using circuit simulation.

There are also quite a bit of variation on how each parameter is measured. It is difficult to disagree on how input-to-output delays are measured, but there is considerable room for setup/hold measurements, input capacitance, static and dynamic power. On such quantities, the algorithms and techniques make a big difference. More often than not, the alternative techniques are wrong or irrelevant to the intended use.

Another false technique used is interpolation. Instead of running the whole sequence of input slope and load points, a few are run and the rest are obtained from interpolation. This is especially true in the face of process variation which requires hundreds or thousands of different libraries. Cutting short of proper characterization steps adds to the process variation and increase it rather than help manage the problem. To address process variation, our YieldOpt product is available without cutting corners.

Accuracy Testing
One of the unique features of LTI tools is the ability to generate automatic test bench circuits, load files and scripts to compare the delays as seen by timing analyzers to delays reported by Spice under the same conditions. This verifies These tests can be performed with lumped capacitive loads as well distributed loads as may be specified by in DSPF format to account for RC effects. CCS and ECSM type delay models can easily verified for accuracy.

"Copyright (c) 2012 Library Technologies, Inc., All Rights Reserved."