Method and apparatus for enhancing the performance of event driven dynamic simulation of digital circuits based on netlist partitioning techniques

The present invention relates generally to the functional verification of electronic designs and more particularly to the partitioning of a design under verification for the integration of dynamic simulation and static timing analysis methodologies. Today's engineers are faced with an increasing difficult task of handling the verification of state-of-the-art system-on-chip (SoC) designs. The various verification techniques in use fall into two major categories, namely the dynamic event-driven simulation and the static timing analysis (STA) techniques. STA techniques are based on simplifying the general model of event-driven computation to that of a synchronous model. By taking advantage of the separation of the timing and functional behavior made possible by the synchronous design style, STA tools can apply complete, rigorous, and efficient algorithms that result in an overwhelming performance advantage when compared to event-driven simulation.

However, the STA techniques are not directly applicable to most designs at the full-chip level. This is because most designs are a combination of synchronous logic blocks and asynchronous logic blocks, or other non-synchronous design blocks, such as embedded analog blocks. One illustration of such a full-chip design 100 is shown in FIG. 1 . This exemplary full-chip design consists of four clock domains: block 102 is driven by CLK 1 ; block 104 is driven by CLK 2 ; block 106 is driven by CLK 3 ; and block 108 is driven by CLK 4 . In addition, the design also contains an asynchronous block 110 and an analog block 112 . As shown in FIG. 1 , a typical design at the full chip level violates the separation of timing and functionality implicit in the synchronous model. Although most of the modules might be synchronous, the clocking methodology that has four different clock domains can result in asynchronous interaction with each other. Furthermore, the design could contain non-synchronous design constructs 110 and 112 that cannot be verified by STA techniques.

As a result, both sets of solutions have their unique problems when applied to the verification of a complete design. On one hand, the static verification algorithms require strict adherence to the synchronous design style. On the other hand, the dynamic event-driven simulation is limited by the computing power and memory capacity of the computers used. Therefore, it would be advantageous to have an improved method for full chip level verification.

Disclosed is a full-chip level verification methodology that combines static timing analysis techniques with dynamic event-driven simulation. The specification discloses capabilities to partition a multiple-clock design into various clock domains and surrounding asynchronous regions automatically and to apply timing behavior during simulation on an instance by instance basis.

Static timing analysis techniques can be leveraged to verify the synchronous cores of each clock domain. The asynchronous regions of the design and the interaction between synchronous cores of the clock domains are validated using detailed dynamic event-driven simulation without the burden of carrying the interior timing attributes of the synchronous cores that have already been verified. With the unnecessary interior timing attributes of synchronous cores removed during dynamic simulation, the disclosed method accelerates the verification process and requires less computing power and memory capacities to complete the verification of the full chip.

The accompanying drawings are included to provide a further understanding of the invention and, together with the detailed description of the preferred embodiment, serve to explain the principles of the invention.

FIG. 1 depicts an illustration of a fill-chip design.

FIG. 2 depicts one approach for determining clock domains.

FIG. 3 depicts one approach for crossing between two clock domains.

FIG. 4 depicts one approach for propagating color through a combinational instance.

FIG. 5 depicts one approach for propagating color through a Flip-Flop instance.

FIG. 6 depicts one approach for propagating color through a latch instance.

FIG. 7 depicts one approach for determining timing for a combinational instance.

FIG. 8 depicts one approach for determining timing for a clocked sequential instance with no asynchronous reset.

FIG. 9 depicts one approach for determining timing for a clocked sequential instance with asynchronous reset.

The disclosed methodology leverages the fact that the synchronous parts of a design have been validated by STA techniques to accelerate full-chip simulation by removing the timing from their synchronous cores. Synchronous cores are determined by netlist partitioning that can be applied at the module level or at the full-chip level. Partitioning at the module level preserves the intended design hierarchy and is a natural extension of the exiting STA techniques. However, when partitioning is applied at the full-chip level, larger synchronous cores are defined since they can include the inter-block regions separating two synchronous blocks controlled by the same clock. Furthermore, the synchronous regions determined by full-chip partitioning are delimited by sequential elements. After partitioning, the timing of the synchronous regions of each clock is verified by STA tools. Having verified the timing of the synchronous cores using STA techniques and generated the corresponding assertion for assumption verification, the design is now ready for full-chip dynamic simulation. Effective full chip verification is achieved by providing simulation acceleration on the synchronous cores and using assertions to validate the STA assumptions.

Detail timing removal from the synchronous core simulation is possible because the core timing have been verified by STA tools. Indeed, timing within the synchronous core is determined by the clocking of the storage elements and not by timing delay propagation. Thus, each sequential stage can be viewed as a static network that is evaluated once for every clock edge. The combinational logic evaluation step can then be collapsed into a single cycle triggering cycle optimizations such as compiling zero-delay logic cones. Eliminating output scheduling also reduces simulation overhead. Timing removal enables efficient full-chip simulation. The simulation can be focused on the interactions between clock domains and the non-synchronous regions of the design such as asynchronous and analog constructs. Hence, timing removal can result in dramatically faster simulation runtimes as well as potentially higher simulation capacity.

The disclosed design partitioning methodology divides all the nets of a full-chip design into regions that should be simulated with Full-Timing, and regions that could be simulated with timing removal. The method comprises three main phases: 1) Clock Domain Determination phase; 2) Netlist Coloring phase; and 3) Timing Determination phase.

A clock domain is defined to contain all sequential elements controlled by a single clock signal together with the combination logic driven by these elements. Determining clock domain is achieved by recognizing exclusivity regions that isolate the domain from the rest of the design. Exclusivity regions stop the flow of data into and out of the domain. The two primary elements that are capable of stopping the flow of data are Flip-Flops and Latches. A Flip-Flop or edge-triggered device is “self exclusive” or able to avoid data seepage by its very nature. Latches are level sensitive devices that are open when their controlling clock is active. In a latch-based design, exclusivity is achieved when closed or inactive Latches stop data flow. This is usually implemented using a multiple-phased clocking scheme. Because these clock phases never occur at the same time, the Latches in an exclusivity region are never open at the same time, thus stopping the flow of data at the boundary of the clock domain. Indeed a Flip-Flop can be thought of as a combination of two Latches: a master and a slave that are always enabled during different phases of the driving clock.

On the other hand, to detect exclusivity regions in a latch-based design, it is important to understand the relationships between co-operating clocks. A number of issues are tracked such as the ‘active’ phase of the Latches, the relationship of all clocks, as well as the effect that a clock path might have on a latch's active phase.

FIG. 2 illustrates one approach to determine clock domains among various circuit elements. Determining the clock domains of a design comprises the step of identification of sequential elements and followed by the step of Clock Walking. In the step of identification of Sequential Elements, the cells in the design are analyzed to determine which cell is sequential in nature. In FIG. 2 , the sequential elements such as the Flip-Flops 202 and 204 are identified in this step. The clock-ports 206 and 208 as well as any set or reset ports of each sequential cell are also identified.

In the Clock Walking step, the list of sequential elements and a user-defined list of input clock signals are traversed starting at the clock pins. The Clock Walking process consists of tracing the loads of user-defined clock signals progressively, until all sequential elements controlled by the given clock signal are determined. The effect of the clock path on the clock signal is also tracked during Clock Walking to determine a latch's active phase. This is used for the detection of exclusivity regions in latch-based designs.

Sequential elements that remain without a controlling clock signal after the Clock Walking step are potentially part of an asynchronous design construct. They are treated as combinational logic in the Netlist Coloring phase.

A clock domain can further be divided into its synchronous core operating in a predictable step-by-step timing pattern with respect to the clock and the remaining logic where asynchronous interactions with the rest of the design can occur. It follows that the elements within a synchronous region can be simulated with No-Timing (NT) whereas those in the periphery should be run in Full-Timing (FT) mode. The purpose of partitioning is to separate the design elements that must operate at Full-Timing from those elements that can operate with no timing without impacting the functional and timing characteristics of the surrounding logic. Whereas combinational cells are divided into Full-Timing or no timing groups, the sequential cells are further subdivided into groups depending on the enabling or disabling of their capability to verify timing checks (i.e. setup and hold constraints).

A key to applying the rules of partitioning is recognizing domain crossings and exclusivity regions. Domain crossing occurs when a logic element from one clock domain drives a logic element in another clock domain. FIG. 3 illustrates one approach to recognize such a domain crossing, where the CLK 1 domain consists of exemplary Flip-Flop 304 and combinational logic 306 elements. Similarly, the CLK 2 domain consists of exemplary Flip-Flop 310 and combinational logic 312 elements. Electronic data cross the clock domains from the combinational logic element 306 in CLK 1 domain to the Flip-Flop 310 element in CLK 2 domain.

Once the clock domains have been determined, the algorithm moves to the Netlist Coloring phase. The objective of the Netlist Coloring phase is to label each port of every instance in the design by a color, represented by a positive integer number that identifies the controlling signal. There are two types of controlling signals: 1) clocks; and 2) external pin signals.

Each clock is identified by its own color and the Color-Zero is used to identify the signals that should operate in Full-Timing. Initially all external pins of the design are assigned the Color-Zero except for the clock pins. External pin signals can change independently of clocks; therefore they are simulated in Full-Timing to capture accurate interaction of the design with the external world. The colors are then propagated to the rest of the netlist by forward traversal with different coloring procedures applied to combinational instances, Flip-Flop instances and latch instances. During netlist traversal, all the traversed ports are stored on a stack. This stack is cleared each time the propagated color changes when a Flip-Flop or a latch-based exclusivity region is reached. If the traversal reaches a different color domain or an output port, then the ports on the stack are assigned the Color-Zero. This process ensures that all the inter-clock regions as well as the output periphery regions are operating in Full-Timing in order to account for accurate asynchronous timing interactions.

FIG. 4 illustrates one approach to propagate color through a combinational instance. The method begins at block 402 , and thereafter passes to block 404 where a determination is made as to whether or not any input port has a different color than the one being propagated. If this condition is true, then the “Yes” path is taken and the output ports are colored with the Color-Zero (block 406 ). Alternatively, if the condition is false, then the “No” path is taken and the output ports of the cell are colored with the color being propagated (block 408 ). In the case that at least two asynchronous signals interact at an instance, the outputs will be labeled with the Color-Zero and therefore the Full-Timing simulation of this instance is used.

This discussion of partitioning and coloring includes examples of embodiments that are in the presence of set and reset ports. Both port types have similar characteristics and effects on the partitioning procedure. The discussion may also include examples of embodiments in the presence of reset ports, however, the partitioning and coloring is also valid for embodiments that include the presence of set port.

FIG. 5 illustrates one approach to propagate color through a Flip-Flop instance. A Flip-Flop is an edge-triggered sequential device that is controlled by the rising or falling edge of the enabling clock signal. The methodology distinguishes between Flip-Flops with asynchronous reset; and those with synchronous reset or no reset i.e. synchronous Flip-Flops. The method begins at block 502 , and thereafter passes to block 504 where a determination is made as to whether an asynchronous reset port exists. If an asynchronous reset port does not exist, i.e. the Flip-Flop is synchronous, the “No” path is taken and the output will be labeled with the same color as the clock (block 506 ), regardless of the color of the other input data signals since output changes are affected the clock signal transitions only. But if the Flip-Flop has an asynchronous reset port, changes in the output signals can occur independently of the clock when the asynchronous reset is activated. Then the “Yes” path is taken and the method continues at block 508 where a second determination is made as to whether or not the clock and reset ports have the same color. If both clock and reset ports have the same color, the output will be labeled with that same color (block 510 ). But if the colors of clock and reset ports are different, the output will be labeled with Color-Zero (block 512 , indicating Full-Timing simulation is used).

FIG. 6 illustrates one approach to propagate color through a latch instance. A latch is a level sensitive device that is open when it controlling signal is active. While a Flip-Flop is self-exclusive by it very nature, a latch has to be controlled with clock signals that have the proper phase relationships to build an exclusivity region. In a latch-based design, inactive clock phases of a latch interrupt the flow of data. To detect a latch-based exclusivity region, the active phases of the latch and the relationship of all the co-operating clock phases have to be tracked. A latch can also have synchronous or asynchronous reset pins. A latch output color depends on whether an exclusivity region has been detected. If an exclusivity region has not been detected, then the data signal transition can affect the output signal when the clock is active. In this situation the three signals, namely data input, clock, and reset are taken into consideration when coloring the output. As shown in FIG. 6 , the method begins at block 602 , and thereafter passes to block 604 where a first determination is made as to whether an asynchronous reset port exists. If there is asynchronous reset port, then the “Yes” path is taken and the method continues at block 606 where a second determination is made as to whether the clock and reset ports have the same color. If the color and reset ports have different colors, then the output ports is colored with Color-Zero (block 608 ). But if an asynchronous reset port does not exist or the color and reset ports have the same color, then the method continues at block 610 where a third determination is made as to whether there is an exclusivity region. If there is an exclusivity region, then the output ports are colored with the color of the clock (block 612 ). Alternatively, the method moves to block 614 where the latch instance is treated the same as a combinational instance as described in FIG. 4 .

In one exemplary approach, the event driven simulator has the capability of simulating a cell instance in one of the four possible timing modes on an instance-by-instance basis. The four timing behavior modes for a cell instance in this exemplary design are:

Full-Timing mode: Complete timing simulation is done. No Timing mode: All timing delays and timing checks are removed. Timing Checks mode (for sequential cells only): Input timing checks are performed. I/O Delays mode (for sequential cells only): I/O path delays are included. After the Netlist Coloring phase, the mode of timing simulation is determined based on the color of the inputs and outputs of an individual instance. The following examples are used to identify the timing status of a cell in a design.

FIG. 7 illustrates one approach for determining timing for a combinational cell instance. The method starts at block 700 and thereafter passes to block 702 . A determination is made in block 702 as to whether or not any port has Color-Zero. If any port is labeled with a Color-Zero, this instance is to be simulated in the Full-Timing mode (block 704 ). If no port has Color-Zero, then the “No” path is taken and the method continues at block 706 where a second determination is made as to whether or not all ports have the same color. If there are any two ports have different colors, then this instance is simulated in the Full-Timing mode (block 710 ). Otherwise, if all the ports have the same color, then this instance is simulated in the No Timing mode (block 708 ).

FIG. 8 illustrates one approach for determining timing for a clocked sequential instance with no asynchronous reset. This method starts at block 800 and thereafter passes to block 802 . A determination is made in block 802 as to whether or not all input ports have same clock color. If all input ports have same clock color, then the “Yes” path is taken and the method moves to block 804 . A second determination is made in block 804 as to whether or not all output ports have same clock color. If all output ports have same clock color, then this instance is simulated in the No Timing mode (block 806 ). Alternatively, this instance is simulated in the IO Delay mode (block 808 ). But if the determination in block 802 turns out that all input ports don't have same clock color, then the method moves on to block 810 where another determination is made as to whether or not all output ports have same clock color. If all output ports have same clock color, then the instance is simulated in the Timing Check mode (block 812 ). Otherwise, the instance is simulated in the Full-Timing mode (block 814 ).

FIG. 9 illustrates one approach for determining timing for a clocked sequential instance with asynchronous reset. This method begins at block 900 , and thereafter moves to block 902 . A determination is made in block 902 as to whether or not the clock and reset ports have Color-Zero. If the clock or reset ports have Color-Zero, then the instance is simulated in Full-Timing mode (block 904 ). Otherwise, the “No” path is taken and the method continues at block 906 where a second determination is made as to whether or not the clock and reset ports have different colors. If the clock and reset ports have different colors, then the instance is simulated in the Full-Timing mode (block 908 ). Otherwise, the timing of this instance is further determined in the same manner as if there is no asynchronous reset port as illustrated in FIG. 8 (block 910 ).

At this point, it should be noted that although the invention has been described with reference to specific embodiments, it should not be construed to be so limited. Those of ordinary skill in the art can modify the claimed invention with the benefit of this disclosure without departing from the spirit of the invention. For example, modifications can be made to introduce minor variations of the timing behavior modes for simulation. In addition, different coloring schemes and formats of color representations can be employed with the disclosed invention. All of these modifications can be applied to achieve desired functional verification goals of electronic designs. These and other uses and modifications are within the spirit and scope of the present invention. Thus, the invention should not be limited by the specific examples used to illustrate it but only by the scope of the appended claim