Test Ready Compiler for Design for Test Synthesis

(1) Field of the Invention

The present invention relates generally to the field of logic synthesis for integrated circuit devices. More particularly, aspects of the present invention relate to an HDL compiler and a scan insertion process used for test implementations designed for circuit synthesis.

(2) Background of the Related Art

Complex integrated circuits are designed with the use of computer aided design (CAD) tools. Specifically, application specific integrated circuits (ASICs) and field programmable gate array (FPGA) circuits can be designed using a variety of CAD tools. The development of ASICs and FPGA circuits with the aid of CAD tools is referred to as electronic design automatic or EDA. Design, checking and testing of large scale integrated circuits are so complex that the use of programmed computer systems are required for realization of normal circuits. This is partly because the integrated devices are inherently complex and partly because the circuit design needs to be decomposed into simpler functions which are recognized by the CAD tool. It is also partly because considerable computation is required in order to achieve an efficient layout of the resultant network. The result of the computerized design process is a detailed specification defining a complex integrated circuit in terms of a particular technology. This specification can be regarded as a template for the fabrication of the physical embodiment of the integrated circuit using transistors, routing resources, etc.

Integrated circuit designs can be represented in different levels of abstraction, such as the register transfer level (RTL) and the logical level, using a hardware description language (HDL), also called high level design language. Two exemplary forms of HDL are Verilog and VHDL. The integrated circuit can be represented by different layers of abstractions (e.g., behavioral levels, structural levels and gate levels). An RTL level is an intermediary level of abstraction between the behavioral and structural levels. HDL descriptions can represent designs of all these levels.

The behavior levels and RTL levels consist generally of descriptions of the circuit expressed with program-like constructs, such as variables, operators conditional loops, procedures and functions. At the logic level, the descriptions of the circuit are expressed with Boolean equations. The HDL can be used along with a set of circuit constraints as an input to a computer implemented compiler (also called a "silicon compiler"). The computer implemented compiler program processes this description of the integrated circuit and generates therefrom a detailed list of logic components and the interconnections between these components. This list is called a "netlist." The components of a netlist can include primitive cells such as full-adders, NAND gates, NOR gates, XOR gates, latches and D-flip flops, etc. and their interconnections used to form a custom design.

In processing the HDL input, the compiler first generates a netlist of generic primitive cells that are technology independent. The compiler then applies a particular cell library to this generic netlist (this process is called mapping) in order to generate a technology dependent mapped netlist. The mapping process converts the logical representation which is independent of technology into a form which is technology dependent. The mapped netlist has recourse to standard circuits, or cells which are available within a cell library forming a pan of the data available to the computer system.

Compiler programs and mapping programs are well known in the an and several of these systems are described in U.S. Pat. No. 5,406,497, by Altheimer et al.

An important pan of the logic synthesis process involves designing for testability. Programs that aid in the testability process of logic synthesis are called design for test (DFT) processes. As part of DFT, it is known to take the mapped netlist generated from a compiler and add and/or replace certain memory cells and associated circuitry with special memory cells that are designed to allow the application of test vectors to certain logic portions of the integrated circuit. The act of applying test vector is called stimulation of the design and the special memory cells and associated circuitry are referred to as DFT implementations. Issues concerning controllability deal with facilitating the application of the test vectors to the circuitry to be tested. The same memory cells can be used to capture the output of the circuitry for observation and compare this output to the expected output in an effort to determine if circuit (e.g., manufacturing) defects are present.

The portions of an integrated circuit that are designed to perform its intended or expected operational function are called its "mission mode" circuitry while the portions added to the integrated circuit to facilitate testability are called "test mode" circuitry or DFT implementations. The resultant circuit therefore has two functional modes, mission and test.

An exemplary flow chart diagram of a typical logic synthesis process, including a DFT process, is shown in FIG. 1. The processes 200 described with respect to this flow chart is implemented within a computer system in a CAD environment. High level design language (HDL) descriptions of the integrated circuit enter at block 201. Also accompanying the HDL 201 is a set of performance constraints 205 applicable to the design which typically include timing, area, power consumption, and other performance related limitations that the compiler 225 will attempt to satisfy when synthesizing the integrated circuit design. Constraints 205 can also include non-performance related constraints such as structural and routing constraints. Compiler 225 consists of a generic compiler 203 (also called an HDL compiler, RTL synthesizer, or architectural optimizer) that inputs the HDL 201 description and generates therefrom a technology independent or "generic" netlist 207 which is also dependent on the constraints 205. As discussed above, the netlist 207 is a list of technology independent components or operators and the interconnections between them.

The generic netlist 207 is then input to a design compiler 209 that includes a computer implemented logic optimization procedure and a mapping procedure which interfaces with a technology dependent cell library 230 (e.g., from LSI, VLSI, TI or Xilinx technologies, etc.). The cell library 230 contains specific information regarding the cells of the specific technology selected such as the cell logic, number of gates, area consumption, power consumption, pin descriptions, etc., for each cell in the library 230. Logic optimization procedure of block 209 includes structuring and flattening procedures. The mapping procedure of block 209 generates a gate level mapped netlist 211 that is technology dependent having cells specifically selected to satisfy the constraints 205. This gate level netlist 211 consists at this point of "mission mode" circuitry.

At block 212 of FIG. 1, DFT process 213 performs a particular test insertion process (here a scan) to implement testability cells or "test mode" cells into the overall integrated circuit design. In this process 213, memory cells of the mapped netlist 211 are replaced with memory cells that are specially designed to apply and observe test vectors or panems to and from portions of the integrated circuit. In one particular DFT process, these memory cells specially designed for test are called scannable memory cells. The test vector patterns can be derived from combinational or sequential automatic test pattern generation (ATPG) processes depending on whether or not a full or partial scan is performed by the scan insertion process 213. Process 213 also performs linking groups of scannable memory cells into scan chains so that the test vectors can be cycled into and out of the integrated circuit design. The output of the scan insertion process 213 is a scannable netlist 215 that contains both mission and test mode circuitry.

A problem occurs in the prior art process of FIG. 1 in that the scan insertion process 213 does not take into account its impact on the mission mode design. Specifically, the addition of the testability cells (scannable cells), and interconnections there between (chaining resources), and the addition of other dedicated connections required for operation of the scan chains (e.g., scan clock routing and scan enable signal routing) can cause the overall design to violate one or more of the defined constraints 205.

Therefore, a second compile process 217 of FIG. 1 (full or incremental compile) is invoked by the prior art process 200 in order to more effectively optimize the scannable netlist 215 to the constraints 205. An incremental compile 217 does not process all existing structure as in a full compile, it only applies high level logical optimization to the unmapped portions of the design. Those unmapped portions are then mapped using a technology dependent library. During a process iteration, an incremental compile 217 always processes to decrease the circuit cost. However, although this second compile process 217 is only an incremental compile process, it applies mapping optimizations iteratively on the entire scatmable netlist 215. As a result, processing time to perform the second compile process 217 can be on the order of weeks given conventional CAD technology and circuit complexity.

Altematively, many prior systems utilize a full compile as the second compile process 217. The full compile process is similar to process 225 in that the full compile process at 217 applies mapping and logic optimizations to the entire design, not just the unmapped portions.

After the second compile process 217 of FIG. 1 completes, a scannable netlist 219 is again generated that contains the testability cells but that may or may not meet the original performance constraints 205. Therefore, at block 221, the prior art then performs a test to determine if the scannable netlist 219 meets the constraints 205. If the netlist 219 meets the constraints, then at block 235, other circuit synthesis procedures continue until the integrated circuit design can be fabricated onto a substrate and tested.

However, as is often the case, the addition of the testability cells by the scan insertion process 213 does not allow the second compile process 217 to meet constraints 205 without a design modification to the original HDL program 201. In such case, the overall process 200 flows from block 221 back to the HDL 201 where the architect modifies the HDL program 201 so that the addition of the testability cells and other resources will eventually satisfy, when possible, the given constraints 205 after the incremental compile step 217 is again executed.

The prior art process 200 of FIG. 1 has several disadvantages. It is disadvantageous to execute a second substantial compile process 217 in an attempt to match the testability cells and linking resources to the given set of constraints. Although this process can be an incremental compile step in that much of the gate level connections are not removed, mapping optimization portions of this compile process still operate in an iterative fashion over the entire design. The addition of this second compile process, using conventional technology, delays the overall integrated circuit synthesis process by as much as one to two weeks. Even after this long delay, there are no guarantees that the incremental compile process 217 will generate a scannable netlist satisfying the constraints 205. In this case, a time consuming task of returning to the HDL for redesign is required. This process involves the chip architect designers once more and, therefore, it is unclear under the prior art system when a designer can sign off on his or her work in the design process.

What is needed is a system that can reduce the time required to perform circuit synthesis while providing effective DFT processes. The present invention offers such advantageous functionality.

Accordingly, it is an object of the present invention to provide a time efficient design synthesis system operable within a computer implemented CAD system that includes effective DFT processes. It is further an object of the present invention to provide the above with the elimination of the second substantial compile process that is used in the prior art. It is yet another object of the present invention to provide the compile process with information relating to the impact of testability cells and resources to the overall design so that this information can be accounted for during the initial compilation phase. It is further an object of the present invention to provide specific aspects of the initial compile process to a modified scan insertion process so that the added testability cells and resources can be processed to likely meet the constraints without requiring a substantial second compilation process. To this end, it is an object of the present invention to provide optimization procedures in a modified scan insertion procedure that operate only on the test cells and resources. These and other objects and advantages of the present invention not specifically recited above will become clear within discussions of the present invention herein.

A computer implemented process and system are described for providing an test ready (TR) compiler with specific information regarding the impact of added scannable test cells and resources on its mission mode design. In so doing, the TR compiler optimizes with increased efficiency for the added test resources so that predetermined performance and design related constraints of the mission mode circuitry are maintained after addition of the test resources. The TR compiler translates generic sequential cells into technology dependent non-scan sequential cells. In the TR compiler, during replacement, scannable memory cells are used in place of these non-scan memory cells specified within the mission mode design and therefore the TR compiler is informed of the characteristics of the scannable memory cells and their links. For test, the scannable memory celth are chained to each other to form scan chains. To account for this chaining during compile, the TR compiler provides output driven loopback connections to simulate the electrical characteristics of this chaining during compile. In the above implementation, the TR compiler can efficiently provide translation of an TR description having test implementations into a gate level netlist. With the addition of certain information regarding the test implementation (e.g., scan replacement is done and loopback connections are added), the TR compiler of the present invention better optimizes the overall design for the addition of the test resources which are linked together for testability.

Specifically, embodiments of the present invention include, in a computer implemented synthesis system, a method of generating a netlist having the computer implemented steps of: receiving an HDL specification representing a design to be realized in physical form and storing the HDL specification in a computer memory unit; receiving constraints applicable to the design; compiling the HDL specification with a compiler to produce a netlist description of the design wherein the netlist comprises functional logic blocks and connections there between including sequential cells and combinational logic, the step of compiling further comprising the steps of: translating generic non-scan cells to technology dependent non-scan cells; replacing the non-scan sequential cells with scannable cells specially adapted to apply test vectors to combinational logic of the design; and installing a loopback connection between an output and an input of each scannable sequential cell; and storing the nefiist description into the computer memory unit, the netlist description including the scannable sequential cells and the loopback connection associated with each scannable sequential cell.

Embodiments of the present invention include the above and wherein the non-scan sequential cells are D flip-flops and wherein the scannable sequential cells are multiplexed input D flip-flops and wherein the step of installing a loopback connection between an output and input of each scannable sequential cell installs a loopback connection between the output and a multiplexed scan-in input of each scannable sequential cell. Embodiments of the present invention further include a system implemented in accordance with the above.

FIG. 1 is a flow diagram illustrating a prior art process for logic synthesis with design for test implementations.

FIG. 2 is an exemplary computer system used in accordance with the present invention as a CAD system for design synthesis.

FIG. 3A is a diagram of a logic model including combinational logic and memory cells used by the present invention to represent the design of a synthesized integrated circuit.

FIG. 3B illustrates replacement performed by the present invention from non-scan memory cells specified by or inferred from the original HDL description into scannable memory cells used for DFT implementation.

FIG. 4 illustrates a procedure of the present invention (Design Rule Checker) that determines valid scan chains and marks as violated any scannable memory cell not pan of a valid scan chain.

FIG. 5A illustrates an exemplary edge triggered D flip-flop used in the HDL description in an embodiment of the present invention.

FIG. 5B illustrates an exemplary scannable memory cell version of the memory cell presented in FIG. 5A according to an embodiment of the present invention.

FIG. 5C illustrates an exemplary edge triggered memory cell used in the HDL description that also contains other combinational logic in addition to the memory cell circuitry.

FIG. 5D illustrates an exemplary scannable memory cell version of the memory cell presented in FIG. 5C according to an embodiment of the present invention including other combinational logic present.

FIG. 6A illustrates an exemplary circuit implementation of a loopback line added by the test ready (TR) compiler of the present invention to simulate a linked scan chain where the loopback is taken from the output (Q).

FIG. 6B illustrates an exemplary circuit implementation of a loopback line added by the TR compiler of the present invention to simulate a linked scan chain where the loopback is taken from the inverted output (/Q).

FIG. 7 illustrates a task performed by the modified scan insertion process of the present invention where the scan chain is buffered in links that span more than one module allowing the loopback lines to accurately simulate this condition.

FIG. 8 is an overall flow diagram of the embodiments of the present invention starting from HDL description to the generation of test vectors having a scannable gate level netlist.

FIG. 9 is a flow diagram illustrating processes of the test ready (TR) compiler of the present invention.

FIG. 10 is an overall flow diagram illustrating processes of the modified scan insertion process of the present invention including the constraint driven compile process.

FIG. 11A and FIG. 11B represent a flow diagram illustrating processes of the constraint driven compile process of the modified scan insertion process of the present invention.

FIG. 12 is a flow diagram illustrating processes of the size design process of the constraint driven compile process of the present invention.

FIG. 13A and FIG. 13B illustrate an exemplary circuit transition performed by the phasing subprocess of the size design process shown in FIG. 12.

FIG. 14A and FIG. 14B illustrate an exemplary circuit transition performed by the buffering subprocess of the size design process shown in FIG. 12.

FIG. 15A and FIG. 15B illustrate an exemplary circuit transition performed by the downsizing subprocess of the size design process shown in FIG. 12.

FIG. 16A and FIG. 16B illustrate an exemplary circuit transition performed by the isolation subprocess of the size design process shown in FIG. 12.

FIG. 17A and FIG. 17B illustrate an exemplary circuit transition performed by the offloading subprocess of the size design process shown in FIG. 12.

FIG. 18A and FIG. 18B illustrate an exemplary circuit transition performed by the balancing subprocess of the size design process shown in FIG. 12.

FIG. 19A and FIG. 19B illustrate an exemplary circuit transition performed by the splitting subprocess of the size design process shown in FIG. 12.

FIG. 20 illustrates the present invention test ready (TR) compiler and modified scan insertion process within a hierarchical design illustrating the practical nature of the present invention on a chip level netlist.

In the following detailed description of the present invention numerous specific details are set forth in order to provide a thorough understanding of the present invention. However, it will be obvious to one skilled in the art that the present invention may be practiced without these specific details. In other instances well known processes, methods, procedures, components, and circuits have not been described in detail as not to unnecessarily obscure aspects of the present invention.

Some portions of the detailed descriptions which follow are presented in terms of procedures, processes, and symbolic representations of operations on data bits within a computer memory. These procedure descriptions and representations are the means used by those skilled in the data processing arts to most effectively convey the substance of their work to others skilled in the art. A procedure, process, or logic block is here, and generally, conceived to be a self-consistent sequence of steps leading to a desired result. The steps are those requiting physical manipulations of physical quantities. Usually, though not necessarily, these quantities take the form of electrical or magnetic signals capable of being stored, transferred, combined, compared, and otherwise manipulated. It has proven convenient at times, principally for reasons of common usage, to refer to these signals as bits, values, elements, symbols, characters, terms, numbers, or the like.

It should be borne in mind, however, that all of these and similar terms are to be associated with the appropriate physical quantities and are merely convenient labels applied to these quantities. Unless specifically stated otherwise as apparent from the following discussions, it is appreciated that throughout the present invention, discussions utilizing terms such as "processing" or "computing" or "calculating" or "determining" or "displaying" or executing a procedure or the like, refer to the action and processes of a computer system, or similar electronic computing device, that manipulates and transforms data represented as physical (electronic) quantities within the computer system's registers and memories into other data similarly represented as physical quantities within the computer system memories or registers or other such information storage, transmission or display devices.

Specific aspects of the present invention are operable within a programmed computer aided design (CAD) system. A CAD system operable to implement the elements of the present invention is shown in FIG. 2. In general, the CAD system of the present invention includes a computer system 112 which includes a bus 100 for communicating information including address, data, and control signals, a central processor 101 coupled with the bus 100 for processing information and instructions, a random access memory 102 coupled with the bus 100 for storing information and instructions for the central processor 101, a read only memory 103 coupled with the bus 100 for storing static information and instructions for the processor 101, a data storage device 104 such as a magnetic or optical disk and disk drive coupled with the bus 100 for storing information and instructions, a display device 105 coupled to the bus 100 for displaying information to the computer user, an alphanumeric input device 106 including alphanumeric and function keys coupled to the bus 100 for communicating information and command selections to the central processor 101, a cursor control device 107 coupled to the bus for communicating user input information and command selections to the central processor 101, and a signal generating device 108 coupled to the bus 100 for communicating signals that are input and output from the system 112.

Program instructions executed by the CAD system can be stored in RAM 102, ROM 103, or in the storage device 104 and when executed in a group can be referred to as logic blocks or procedures. It is appreciated that data produced at the various logic synthesis stages of the present invention, including representations of the different levels of abstraction of the integrated circuit design, can also be stored in RAM 102, ROM 103 or the storage device 104 as shown in FIG. 2.

The display device 105 of FIG. 2 utilized with the computer system 112 of the present invention may be a liquid crystal device, cathode ray tube, or other display device suitable for creating graphic images and alphanumeric characters recognizable to the user. The cursor control device 107 allows the computer user to dynamically signal the two dimensional movement of a visible pointer on a display screen of the display device 105. Many implementations of the cursor control device are known in the art including a trackball, mouse, joystick or special keys on the alphanumeric input device 105 capable of signaling movement of a given direction or manner of displacement.

FIG. 3A illustrates a circuit model 300 utilized by the present invention to represent a logic unit of an integrated circuit. The model 300 includes a memory cell block 301 that outputs signals over line 311 to a combinational logic block 305. Combinational logic 305 also outputs signals over line 313 to drive inputs of the memory cells 301. Memory 301 receives a primary input signal 307 which typically originates off chip. Combinational logic block 305 also receives a primary input signal 309 which typically originates off chip. Memory block 301 generates a primary output signal 315 that goes off chip and combinational logic block 305 also generates a primary output signal 317 that goes off chip. According to model 300, memory block 301 is composed either of edge sensitive clocked memory cells (e.g., flip-flops) or level sensitive memory cells (e.g., latches or registers) or can be composed of other cell types. A variety of memory modes or styles, including either of the memory modes described above, are acceptable within the present invention. For purposes of explanation, the memory cell style adopted within discussions herein is the edge triggered flip-flop but the present invention is equally adapted for other styles such as level sensitive modes, e.g., Level Sensitive Scan Design (LSSD) modes, clocked scan modes, clocked LSSD modes, and auxiliary clocked LSSD modes.

As shown by FIG. 3A, portions of combinational logic 305 that are directly stimulated by primary input 309 and that are direcfiy coupled by its output to primary line 317, can be readily tested for faults (faulty circuitry) by direct application of test vectors (predetermined data patterns) over line 309 and by direct observation of the output over line 317. However, this represents only a small percentage of the logic of block 305. In typical applications, most of the logic gates within block 305 receive their inputs from memory cells within 301 and forward their outputs to other memory cells within 301. In order to accurately test the combinational logic within block 305, DFT processes provide a mechanism for isolating different logic groups within block 305 by (1) passing test vectors into the memory cells 301, (2) allowing the stimulated logic groups to store the product information in predetermined memory cells, and then (3) recalling the output from stimulated portions of block 305 from memory 301.

In effect, test vectors are scanned into the memory cells during test mode, the combinational logic is operated and its output is then captured in the memory cells to be scanned out.

FIG. 3B illustrates a portion of the structure of the mapped memory cells within unit 301 as typically specified or inferred in an HDL circuit description without the addition of test circuitry; these cells are called non-scan cells. Unit 301 consists of a plurality of individual non-scan memory cells. In this example, five exemplary D flip-flops are illustrated 307a-307e. Each non-scan memory cell 307a-307e receives an input 306a-306e from either a primary input off chip, from the combinational logic block 305, or from another memory cell within block 301. These non-scan memory cells also have outputs 309a-309e that typically drive combinational logic gates or other memory cells or can be a primary output that goes off chip.

Within one embodiment, a scan replacement process of the present invention replaces the non-scan memory cells 307a-e of unit 301 with scannable memory cells 320a-e (FIG. 3B) that are provided for DFT functionality and links them together to form a scan chain. A scannable memory cell is also called a test memory cell. Memory cells are also called sequential cells. The result is shown in memory unit 320 of FIG. 3B which contains scannable memory cells 320a-e in a scan chain configuration. This configuration 320 accommodates testability for stuck at faults and allows loading of test vectors into the cells and scanning of certain data out of the cells in response to the application of the test vectors.

Herein, a non-scan or unscan memory cell indicates a memory cell that does not support scanning, e.g., cell elements 307a-e of FIG. 3B. A scan cell or scannable cell indicates a memory cell that supports scanning, e.g., cells 320a-e of FIG. 3B. An unscannable or nonscannable memory cell indicates a memory cell that is violated or otherwise user indicated as not to be scanned.

The scannable memory cells 320a-e in this example consist of multiplexed input D flip-flops and are linked together in chains, as shown, to form shift register configurations. Each particular scannable memory cell can be analogous to each other cell, so one cell 320a is described herein. Scannable cell 320a contains a memory cell 321a and a multiplexer 323. The D input of cell 321a is coupled to the output of a multiplexer 323 which has a select line input 325, called the scan enable or SE line. The data inputs to the mux 323 are (1) an I input 327 analogous to input 306a from unit 301 and (2) an SI shift input 329 which originates from a previous scannable memory cell or from a primary input provided cell 320a is the first cell in a scan chain. It is appreciated that the output of the mission mode logic 305a is typically routed (e.g., by a designer) to the I inputs 327 of the scannable memory cells for observation or to a primary output. The output 331 of cell 321a is muted to mission mode circuitry 305a of the combinational logic 305 and is also muted to another scannable memory cell (e.g., 320b) or to a primary output, if this cell is the last cell of a scan chain. It is appreciated that the Q or /Q output pin can be utilized in the chaining configuration.

In this configuration, scannable memory cells 320a-e comprise an exemplary scan chain because the output of one sequential cell is coupled to the input of an adjacent sequential cell.

The circuitry shown in unit 320 has two modes, mission and test. In mission mode, the SE lines are not asserted and data is selected by the mux from the I inputs. In test mode, the SE inputs of each scannable memory cell 320a-e are asserted such that the shift inputs SI are active and a test vector (string of bits) can be inserted into the integrated circuit through a primary input and shifted through the scannable memory cells 320a-e and then applied to the appropriate logic 305a-e coupled thereto. The product information generated by the tested logic (305a-e in this example) is then stored in a scan chain and shifted out to a primary output for observation. In this manner, the testability of the combinational logic block 305 is greatly enhanced as logic essentially buffed deep within a pipeline of the integrated circuit (e.g., coupled only to scannable memory cells) can be directly isolated and tested by shifting test vectors into a scarunable memory cell chain and then shifting the product back out to a scannable memory cell chain for observation.

Not all scannable memory cells can be included within a scan chain for a variety of reasons. Some memory cells do not offer a scan-in ability or do not offer a scan-out ability due to logic considerations. Further, some cells cannot capture data at their data in port. Also, some memory cells directly control the reset, clock or input line of another memory cell in a scan chain. In this case, when the test vector is loaded into the scan chain, portions of the test vector can be altered, e.g., where the test vector data loaded into a first scan cell is used as a set or reset signal applied to a second scan cell thereby overwriting the test data in the second scan cell of the scan chain. Lastly, some memory cells, after the scan chain is constructed, do not offer scannability for various logic reasons, e.g., the clock input is not clocked properly.

Memory cells that cannot be placed into a scan chain are violated by the present invention and removed from its scan chain. FIG. 4 illustrates an exemplary violated memory cell. A sample scan chain as output from constraint driven scan insertion block 645 of the present invention (FIG. 8) is shown including scannable memory cells 320a-320c of memory unit 301. Block 645 is described further below. Assume that logic block 402 of FIG. 4 is a scannable memory cell and directly controls either the clock input or the set/reset input of memory cell 321d. Because memory cell 321d is directly controlled by another sequential element, this memory cell 321d is violated (e.g., eliminated) from the scan chain so that only cells 321a-c remain in the chain. As shown in FIG. 4, the violated cell 321d is not scan replaced. Cell 321c can be the last memory cell of the scan chain or can be coupled to another memory cell (not shown). In either case, the output of cell 321c is not coupled to violated cell 321d.

Although not shown, it is appreciated that combinational logic 305 outputs to primary outputs (off chip) or to scannable memory cells. The present invention utilizes a well known procedure, Design Rule Checking (DRC), to perform the above analysis to check for violated scan cells. No meaningful observations are possible for circuitry coupled to violated memory cells.

An unfortunate result of violating memory cell 321d is that the combinational logic block 305' is no longer able to be stimulated by the DFT circuitry. It is appreciated that if a memory cell is violated that happens to be the observation cell for block 305', instead of being the stimulus cell, this condition also results in block 305' falling out of the DFT circuitry because its results cannot be observed. As discussed further below, the design rule checking (DRC) process of the present invention determines which scans cells to violate of a constructed scan chain and also performs other functions. A violated scan cell is called an unscannable memory cell.

As discussed further below, the computer implemented TR compiler 625 (FIG. 8) of the present invention performs two important functions in predicting the impact of the DFF circuitry on the mission mode circuitry. First, the TR compiler 625 of the present invention replaces the non-scan cells (e.g., HDL specified or inferred memory cells) with scannable memory cells. An HDL specified or inferred memory cell is a non-scan memory cell. This process includes an equivalence test in order to provide the proper scannable memory cell for replacement. Secondly, in order to simulate the presence of the links that connect the scannable. memory cells into chains, the TR compiler 625 of the present invention provides loopback circuits that couple the output (either Q or /Q) of a given scannable memory cell back to the scan input (SI) of the same memory cell. This loopback connection provides the TR compiler 625 of the present invention with enough information to determine the electrical impact of the DFT circuitry on its design without knowing the actual routing of the chain, which will be determined by subsequently driven processes. In this way, the TR compiler 625 of the present invention can generate its design with enough information regarding the DFT circuitry so that the circuit constraints associated with the HDL specification can more likely be satisfied when the DFT circuitry is completed. The loopback connection is discussed in more depth further below.

FIG. 5A illustrates a sample non-scan memory cell. The exemplary cell 415 shown in FIG. 5A is an edge triggered D flip-flop, however, the present invention can readfly operate with a level sensitive latch or register (e.g., LSSD mode). Once the TR compiler 625 of the present invention performs its mapping functions with a particular technology, the TR compiler 625 replaces the HDL memory cell 415 with an equivalent scannable (or "test") memory cell 425 shown in FIG. 5B. As shown in FIG. 5B and discussed previously, a scannable edge triggered memory cell 425 contains a multiplexed input with mux 430 driving the D input. The mux 430 receives two inputs, an SI input 437, which is a scan input, and another input, I, 435, which is a data input. A scan enable select line SE 439 provides the selection control for data or scan input. Using line 439, scan input (SI) is selected during test mode and data input. (I) is selected during the mission mode.

As is often the case, memory cells contain additional circuitry aside from the circuitry that performs their sequential memory functions. FIG. 5C illustrates one exemplary non-scan memory cell 417 that also contains combinational logic 450 in addition to its circuitry to perform its memory functions. The TR compiler 625 of the present invention will account for this combinational logic 450 when determining an equivalent scannable memory cell for replacement. This determination process is called scan equivalence and is described in more detail below. FIG. 5D illustrates an equivalent scannable (or "test") memory cell 427 that contains the combinational logic 450 that was also present in the non-scan memory cell 417 of FIG. 5C as well as the multiplexed input to provide for scan chaining. In this fashion, the scannable memory cell 427 is equivalent to the non-scan memory cell 417 and will be used by the TR compiler 625 of the present invention to replace cell 417.

After the sequential circuitry (e.g., the memory cells) have been replaced with scannable memory cells, the present invention TR compiler 625 provides loopback connections to effectively simulate the scannable memory cells in a scan chain configuration. FIG. 6A illustrates a loopback connection 440 of the present invention from the Q output of scannable memory cell 425 to the scan input SI of mux 430. In operation for DFT purposes, while memory cell 425 will not be so coupled, the loopback connection 440 will accurately simulate the electrical characteristics seen by the Q output port of cell 425 because, in a scan chain, this port will likely be coupled to another scannable memory cell having similar input characteristics as mux 430 and since the compile wire model uses a unit delay model for this connection. Similarly, the loopback connection 440 accurately simulates the electrical characteristics seen by the mux 430 of cell 425 because in its scan chain mux 430 will likely be coupled to another scannable memory cell having similar output characteristics as the Q output of cell 425 and since the compile wire model does not distinguish between inter and intra module connections. As shown in FIG. 6A, the I input of mux 430 can be coupled to other combinational logic 305' over line 435 and the Q and /Q outputs of cell 425 can be coupled to combinational logic 305.

During compilation, the TR compiler 625 of the present invention will therefore accurately operate on this scannable memory cell 425 with the loopback connection so that constraints will more likely be met after the DFT circuitry is complete.

It is appreciated that the loopback connection 440 employed by the TR compiler 625 of the present invention can also originate from the /Q output of cell 425 as shown in FIG. 6B. Here, the loopback connection 440 is coupled from the /Q port of cell 425 to the SI input of mux 430. The selection of Q or /Q will depend on information determined by the TR compiler of the present invention. This information can originate from library attribute information (e.g., indication of a test scan out) associated with the scan cell. It is appreciated that the selection of Q or /Q is not vital to the present invention because logically either output of cell 425 can be utilized for the scan chain. What is important, however, is that at least one output Q or /Q be selected with a loopback connection 440 for optimization. As shown in FIG. 6B, the I input of mux 430 can be coupled to other combinational logic 305' over line 435 and the Q and /Q outputs of cell 425 can be coupled to combinational logic 305.

FIG. 7 illustrates a circuit modification performed by the present invention after the TR compiler 625 is ran. Since scan chains can span between module, this modification occurs when a link of a scan chain spans more than one module. A module represents a design portion that can be separated by some area within the integrated circuit. The TR compiler 625 applies a unit delay associated with each loop back connection 440 which will accurately characterize most scan links that are intra-moduie but will not accurately model scan links that are inter-module. In the latter case, a gating element is added to satisfy the assumption made by the TR compiler 625.

In FIG. 7, the link spans between two memory cells, one in module A (cell 425) and another in module B (not shown). In these cases, the modified scan insertion process (to be discussed below) of the present invention places a gate 510 (e.g., an AND gate) in the link between the modules. One input of the gate 510 is coupled to scan enable (SE) line 439. With the gate 510 inserted in the scan connection; this physically disables the scan path when SE 439 is not asserted. All the scan path sees electrically is the capacitance of the gate input. Within the wire model of the TR compiler 625, a unit delay is assumed for this scan path. Using the configuration of FIG. 7, the loopback connection 440 of the present invention effectively simulates the mission mode characteristics of this link because the gate 510 hides the mission effects of the long scan link between modules A and B as seen by the Q output of cell 425 and is consistent with the assumptions made by the TR compiler 625. Thus, the addition of gate 510 accurately models the unit delay assumption made in the TR compiler 625 for the scan link.

FIG. 8 illustrates an overall flow diagram of a synthesis process 600 in accordance with embodiments of the present invention and its logic blocks are implemented within the computer controlled CAD system described above. Flow 600 includes the TR compiler (test ready compile) 625 of the present invention and the modified scan insertion procedure 645 (constraint driven scan insertion) of the present invention. Flow 600 also includes a generic HDL compiler 615 which will be described below but is different from the TR compiler 625 of the present invention. Generic compiler 615 is analogous to generic compiler 203.

The flow 600 receives an HDL description 605 of an integrated circuit layout along with a set of design constraints 610 (including design rule limitations, and performance limitations, such as area, timing, power, etc.) that are pertinent to the circuit described by the HDL description 605. Design rules as used herein refer to maximum fanout, maximum signal transition time, and maximum node capacitance. The HDL description 605 can be of a number of different formats, such as VHDL or Verilog and can also represent an entire IC design, but typically represents a module of the overall IC design. The HDL description 605 can be stored in a computer memory unit (e.g., unit 102 or 104) and is fed into an optional generic compiler logic block 615 that is well known in the art. This generic compiler 615 transforms the HDL description 605 into a technology independent netlist 620 that is more readily recognized by the TR compiler 625. Block 615 performs a process on the input netlist 620 to generate a technology independent or generic netlist of the IC layout by interfacing with a synthetic library or "designware" library. Technology independent netlist 620 is composed of logical primitives and operators of the IC layout but the components described therein contain no structure. The generic compile process 615 and resultant output 620 are well known in the art.

The netlist 620 of FIG. 8 generated by block 615 is input to the TR compiler logic block 625 of the present invention. The TR compiler 625 is described in more detail in FIG. 9 which illustrates the particular stages within the TR compiler 625 where the sequential circuits are replaced and where the loopback connections are inserted. The TR compiler 625 in FIG. 8 performs a process on the generic netlist that interfaces with a technology specific library, "cell" library, so that a mapped netlist can be generated that includes specific structural information about the components used in the compiled design. In addition, the TR compiler 625 of the present invention: (1) replaces the HDL memory cells specified in or inferred from the netlist 620 (e.g., non-scan cells) with scannable memory cells; and (2) inserts loopback connections in each scannable memory cell added. FIG. 5A, FIG. 5B, FIG. 5C and FIG. 5D illustrate the memory cell replacements performed by the present invention while FIG. 6A and FIG. 6B illustrate the addition of the loopback connections performed by the present invention TR compiler 625.

Referring to FIG. 8, by performing the above processes, the TR compiler 625 of the present invention is able to better optimize for the eventual construction and completion of the DFT circuitry. In this way, it is more likely that the resultant test circuit design will meet constraints 610. The output of the TR compiler 625 of the present invention is a non-scannable technology dependent netlist 630 that comprises scannable memory cells with loopback connections 440. Although called "nonscannable" because of the loopback connections, netlist 630 is nevertheless a fully scanned netlist in that the TR compiler 625 replaced each HDL specified or inferred sequential cell (e.g., non-scan cell) by an equivalent scannable cell. This netlist 630 can be stored in a memory unit of the computer system such as RAM 102 or the storage device 104 (FIG. 2). The overall non-scannable netlist 630 is optimized to constraints 610 and is a gate level mapped netlist and therefore is technology specific.

The non-scannable netlist 630 of FIG. 8 is then input to a DFT design rule checker logic block 635 ("DRC"). Any of a number of well known DRC processes can operate within the present invention including a DRC as described by E. B. Pitty, D. Martin, and H. T. Ma in a paper entitled "A Simulation-Based Protocol-Driven Scan Test Design Rule Checker," published in IEEE International Test Conference, page 999, paper 40.2 (1994). The DRC 635 checks the scannable memory cells in the nefiist 630 to determine which cells should be violated according to the discussion herein with respect to FIG. 4. Processing flows to DRC block 635' where those scannable memory cells that are determined not to be part of a scan chain by DRC block 635 are marked as violated by the DRC at logic block 635'. Cells marked as violated will be unscanned by the modified scan insertion process 645. The act of unscanning replaces the violated memory cells with an equivalent non-scan memory cell. When unscanned, the loopback connection 440 associated with a violated cell is also destroyed. When violated, the memory cell is referred to as unscannable.

Referring to FIG. 8, the output of the DRC block 635' is a non-scannable nefiist (like 630) but with the violated memory cells marked so as to be unscanned. The output of the DRC 635' is input to a modified scan insertion and routing logic block 645 of the present invention which is also called "constraint driven" scan insertion. This procedure 645 is further described in FIG. 10. Referring to FIG. 8, the modified scan insertion and routing procedure 645 of the present invention breaks the loopback connections 440 in the scannable memory cells of nefiist 630. Also, for memory cells that are not violated but remain non-scanned (e.g., the cells are still HDL specified sequential cells because they were output from a prior art compiler), process 645 replaces the non-scan memory cells with scannable memory cells. It is appreciated that the TR compiler 625 of the present invention does not output any non-scan yet not violated memory cells to logic block 645. Logic block 645 also unscans any memory cell marked as violated.

Logic block 645 of FIG. 8 allocates resources to construct the scan chains between the scannable memory cells. Well known methods and procedures for linking scannable memory cells can be used to create the scan chains for DFT. The present invention modified scan insertion process 645 then performs a reduced set of compiler procedures (constraint driven procedures) that are described with reference to FIG. 11A and FIG. 11B in order to efficiently and effectively optimize the resultant circuitry to the original constraints 610. These compiler routines of logic block 645 are different from the incremental compile step of the prior art in that there are different levels of user selectable effort applied to the compilation process, some of which are applied only to the circuitry added for DFT implementation. The resulting process is more efficient and requires substantially less time to process while yielding excellent results in optimizing the overall circuit layout to the constraints.

The output of logic block 645 of FIG. 8 is a mapped scanstable netlist 650 at the gate level. This netlist 650 contains valid scan chains of scannable memory cells used to receive test vectors. The netlist 650 is incorporated, along with other module scannable netlists, into a scannable system netlist 653 by system composition block 652. The system level scannable netlist 653 is then input to an automatic test pattern generation procedure (ATPG) 655. ATPG procedures are well known in the art and are used to generate the test vectors 660. If the netlist 650 contains a full scan then a combinational ATPG process can be used at logic block 655 to generate the test vectors. If the netlist 650 is the result of a partial scan, in which only some of the memory cells are scanned while others are not, then a sequential ATPG process can be executed at block 655 to generate the test vectors. Both the combinational and sequential ATPG processes are well known. Also at block 655, ATPG formatting is performed in which the test vectors generated by the ATPG processes are altered or modified so that they will operate with a particular specified set of test equipment that accepts a particular format as input.

The test vectors 660 generated from block 655 are therefore customized to a particular test equipment protocol. These test vectors 660 can then be used along with the particular test equipment to load test vectors into the result IC chip to check for faults within the chip.

FIG. 8 illustrates two separate embodiments of the present invention. The first embodiment is the sequential element replacement and the addition of the loopback connections in the TR compiler 625 of the present invention. The second embodiment represents the addition of the reduced set of compile steps (e.g., constraint driven) performed by the modified scan insertion process 645 of the present invention. Both of these processes are described in more detail to follow.

FIG. 9 illustrates processes of the computer system 112 implemented TR compiler 625 of the present invention. As will be discussed below, several of the processes (e.g., logic blocks) shown in FIG. 9 are described for completeness in terms of a particular embodiment of the present invention. Several of these processes are optional and, without departing from the scope of the present invention, can be omitted or replaced with well known procedures. The TR compiler 625 of the present invention starts at logic block 710 (optional) where finite state machine (FSM) optimization is performed. At logic block 710, state machine encoding or descriptions are translated into recognizable or standardized HDL primitives. At logic block 710, the compiler 625 processes each FSM in the design and optimizes their state encoding and then converts them to Boolean equations and technology-independent flip-flops.

At optional logic block 715, subdesigns in the HDL are ungrouped to allow merging of HDL from FSM designs with other HDL portions. By default, subdesigns are compiled hierarchically, preserving their design boundaries. However, this optional logic block 715 will ungroup the subdesigns into their parent designs before being compiled by other procedures. At logic binck 720, the present invention TR compiler 625 performs high level optimizations including resource allocation and sharing depending on timing and area considerations. Additional optimizations such as arithmetic optimization and the sharing of common subexpressions are also performed at logic block 720. In resource sharing, when possible, particular operators such as adders, registers, etc. are shared between different expressions.

Referring to FIG. 9, at logic block 725, synthetic library implementation selection is performed. In this process, generic operators are mapped to architectural representations (implementations). Generic operators that can map into more than one architectural generic component are assigned to a paxticular architectural element based on the constraints 610. At block 725, a technology independent designware library can be used to Select to proper architectural representation. The designware library is created by a designware developer.

The implementation used in block 725 is dependent on the defined constraints 610 (FIG. 8) and the results of logic block 725 contain technology independent generic cells. At logic block 730 of FIG. 9, sequential inference is performed by the TR compiler 625 of the present invention. At step 730, a particular memory style or mode can be selected, e.g., it can be determined whether to use edge triggered flip-flops or level sensitive latches for the sequential elements, or to use another mode. At logic block 730, the selected technology-independent memory elements are then inserted into the design.

At logic blocks 735 and 740, the present invention performs two optional and technology independent processes to fiatten and/or add structure to the design. These processes are well known. At logic block 735, an optional flattening step is performed where the logic is reduced in order to eliminate intermediate variables and results. During flattening, intermediate variables and therefore intermediate logic structure is removed from a design. While advantageous, flattening can have an adverse effect on CPU 101 processing time and design area. Table I below gives an example of a design before and after flattening.

At logic block 740, the present invention performs an optional logic block to stmctttre the design. Structuring is an optimization step that adds intermediate variables and logic structure to a design. During structuring, the TR compiler 625 searches for sub-functions that can be factored out, and evaluates these factors based on the size of the factor and number of times the factor appears in the design. The sub-functions that most reduce the logic are turned into intermediate variables and factored out of the design equations. Table II below illustrates results of structuring a set of equations. In this example, the sub-function t0 is isolated as an intermediate variable and then factored out of the remainder of the design.

After logic block 740, the TR compiler 625 of the present invention, at block 750, translates the sequential elements that were inferred by logic block 730. At logic block 750, the memory cells inferred by logic block 730 are translated into technology dependent non-scan memory cells of a particular target technology. As shown, the technology library 743 is coupled to interface with logic block 750 to provide the cell library for translation. At block 752, the present invention performs scan cell replacement where the technology independent non-scan cells translated by block 750 are then replaced with technology dependent scannable memory cells. The translation procedure 750 performed by the present invention is done with knowledge regarding the types of scannable cells available within the target technology library 743 so that technology dependent non-scan cells inserted by process 750 can be replaced with equivalent scan cells by replacement process 752.

The process of replacement (step 752) involves an equivalence determination because, as shown in FIG. 5C and FIG. 5D, the sequential memory cells can include additional circuitry that needs to be matched during the translation. To perform the equivalence, each component in the technology library 743 contains an ASCII function identification string ("function identifier") that describes in a particular format which functions are implemented in any particular device.

The technology dependent non-scan memory cells placed into the circuit at logic block 750 of FIG. 9 contain similar ASCII function IDs. The present invention at logic block 752 utilizes the function IDs of the non-scan memory cells to compare against function IDs of the scannable memory cell elements of the target technology library 743 to locate scannable memory cells with equivalent functionality as the non-scan memory cell. In this way, functionally equivalent scannable memory cells can be used to effectively replace the non-scan memory cells. Technology libraries 743 include timing information (e.g., set up and hold timing) that is used by the present invention at block 752 to arbitrate between functionally equivalent scan cells for replacement. It is appreciated that TR compiler 625 can also use sequential mapping as an alternative to function ID-based equivalence during replacement 752. In accordance with the present invention, once the scannable memory cells are in place, the resultant design represented in a netlist description can be stored in RAM 102 or storage device 104 of the computer system 112 or any media storage unit.

It is appreciated that in an alternate embodiment of the present invention, logic blocks 750 and 752 can be combined into one replacement step wherein the HDL specified generic sequential cells are directly replaced by technology dependent scannable sequential cells.

At logic block 755 of FIG. 9, the TR compiler 625 of the present invention then adds the loopback connections 440 which connect an output of each scannable memory cell to the scan input of the same cell to simulate a link in a scan chain as shown in FIG. 6A and FIG. 6B. This is performed for each scannable memory cell. In the case of D-flip flop cells, the loopback connection 440 can connect from the Q or /Q output to the preferred scan input (SI). In accordance with the present invention, once the scannable memory cells have loopback connections 440 installed, the resultant design represented as a netlist description can be stored in RAM 102 or storage device 104 of the computer system 112 or any media storage unit.

At logic block 760, the TR compiler 625 then performs a second optional flatting procedure analogous to logic block 735. The second optional flattening procedure is used to perform flattening, if desired, given knowledge of the electrical characteristics of the elements added by blocks 750, 752 and 755 of the present invention. This additional knowledge provides a more refined flattening output. At logic block 765, the TR compiler 625 then performs a second optional structure adding procedure analogous to logic block 740. The second optional structure adding procedure is used to add structure, if desired, given knowledge of the electrical characteristics of the elements added by blocks 750, 752 and 755 of the present invention.

At logic block 770 of FIG. 9 an initial combinational mapping step is performed by the present invention where the generic cells ("architectural implementations") of the design are replaced with technology specific components and more specific timing information and area usage can be obtained. Logic block 770 is linked with the selected technology library 743 (also called a "cell library") defining structure and function of particular elements and components that can be used to replace the generic cells of the design. During block 770, the first functionally equivalent library cell is selected. At the completion of block 770, a technology dependent netlist is generated.

At logic block 775 of FIG. 9, the present invention TR compiler 625 performs input/output pad insertion and optimization. I/O pad insertion is the process of adding I/O buffers to primary inputs and outputs of a design. The characteristics of instantiated I/O pads can be predefined, such as voltage levels, current levels, pull up and pull down resistors, slew rate control, and so forth. I/O pad optimization modifies FO pads that were previously inserted to meet constraints 610. At logic block 775, I/O pads can be optimized to include sizing and incorporation of core logic functionality into the pads. Any of a number of well known procedures can be used at this step.

At logic block 785 of FIG. 9, the present invention TR compiler 625 then performs mapping optimization. The goal of mapping optimization 785 is a gate level implementation that meets the constraints 610 defined in FIG. 8 (e.g., timing, porosity, area, power consumption, etc.). It is applied after the final implementation of the sequential logic is determined. Specifically, logic block 785 of the present invention compares each of its circuit implementations against the constraints 610 to gauge the overall quality of that implementation. Constraints 610 determine which transformations are accepted and which are rejected. There are a number of processes that operate at this step in order to optimize the overall design to satisfy the design rule and performance constraints 610. In general, logic block 785 identifies critical design points throughout the entire design and applies optimization techniques to these points in an auempt to satisfy the constraints 610. A critical point is a point that does not satisfy determined constraints. A critical load is a load associated with a critical path that does not meet determined constraints. A noncritical load a load associated with a path that satisfy determined constraints.

Logic block 785 is applied in an iterative basis over the entire design in an attempt to satisfy constraints 610. During this iterative approach, logic block 785 contains a heuristic that can apply optimizations that locally increase circuit cost, but over the length of the process can reduce circuit cost when applied iteratively over the entire design. Some of the processing of optimization logic block 785 applied at critical circuit points includes: sizing drivers and loads; phasing; buffering; downsizing; isolation; offioading; balancing; and splitting. These are described in more detail with respect to FIG. 12. One embodiment of the present invention advantageously utilizes some of the optimization features of logic block 785 in the modified scan insertion process 645 which is described further below.

At logic block 790, the present invention TR compiler 625 performs logic verification to determine if the design generated by logic block 785 and the HDL design input at logic block 710 are functionally equivalent. Timing considerations are ignored for logic verification. In logic block 790, Boolean matching is performed between the two designs at specified points within the logic design. Logic block 790 can also be used to check manual implementations or changes against the HDL specification. The processing of the TR compiler 625 then exits.

The output generated by the TR compiler 625 of the present invention (see FIG. 8) is a non-scannable netlist description 630 of the design. This netlist description 630, which can be stored in a computer memory or storage unit, contains scannable memory cells (scannable sequential cells) with loopback connections 440 associated with each scannable memory cell.

It is appreciated that the netlist 630 is non-scannable because the loopback connections 440 do not form proper scan chains. However, since each sequential cell has been translated at block 750, and scan replaced at block 752, although the netlist is non-scannable it is nevertheless fully scanned.

An embodiment of the present invention includes a modified scan insertion logic block 645 that performs a modified version of compile (referred to as "constraint driven" compile) that is applied to the translated and replaced sequential cells added by the TR compiler 625 of the present invention. By selectively applying portions of compile routines in a particular fashion during the scan insertion process, this embodiment of the present invention avoids the time consuming incremental compile process required of the prior art. Instead, the modified scan insertion process 645 provides a tiered effort "constraint driven" compile routine that provides effective results with significantly reduced process time.

Further, by application of specific optimization murines during the scan insertion phase, the present invention offers increased flexibility at this phase because logic changes can be more readily made during scan insertion that alter the gate level connections. These optimizations are discussed below.

Specifically, the modified scan insertion block 645 of FIG. 8 is described in more detail with reference to FIG. 10. The logic blocks of FIG. 10 are implemented within computer system 112. Unlike the present invention TR compiler 625, prior art compilers do not provide scannable sequential cell translation. Therefore, the modified scan insertion logic block 645 of the present invention provides logic for processing (1) netlists that have scannable sequential cells and (2) those netlists that do not have scannable sequential cells so that block 645 is compatible with both compiler types.

With reference to FIG. 10, logic block 645 begins at logic block 805 which receives a netlist output by the DRC 635 (FIG. 8). This netlist can contain sequential cells that are marked as violated. Block 805 unscans those sequential cells that are marked as violated or marked as otherwise "do not scan." These cells are unscannable. This is essentially a reverse of the replace sequential cells block 752 and at block 805 the scannable sequential cell is replaced by a timing and functionally equivalent unscannable sequential cell.

It is appreciated that block 645 can also receive input from a prior art compiler which does not perform scannable sequential translation. In such case, the input netlist can contain non-scan memory cells (e.g., they are not scannable sequential cells). At block 810, the present invention identifies those non-violated sequential cells that are not scan replaced. The present invention modified scan insertion block 645 then scan replaces these identified sequential cells with equivalent technology dependent scannable memory cells.

The process of sequential cell replacement applied at block 810 of FIG. 10 is analogous to the sequential replacement block 752 of FIG. 9 that can utilize a function identification string for comparison. If the above string identification approach fails, the present invention applies sequential mapping. Sequential mapping is performed at block 810 based on design rules and area constraints while performance constraints are not considered.

At logic block 815 of FIG. 10, the present invention modified scan insertion process 645 determines which cell output to select as appropriate to couple a particular scannable sequential cell to another scannable sequential cell in a valid scan chain. Assuming an exemplary D tip-flop model, block 815 selects the Q output, or the /Q output, an equal cell output, or opposite cell output. This selection is based on rules that attempt to best optimize to the constraints 610. With equal cells outputs, the output having the most slack (e.g., most favorable timing characteristics) is selected to be the scan-out driver to better optimize the design to timing constraints. In general, at block 815, the present invention atempts to select the output that will least likely impact the mission mode circuitry of the overall design. In other words, if one output (e.g., Q) is coupled to critical mission mode combinational logic, the present invention will attempt to connect this sequential cell to the scan chain using the other output (e.g., /Q).

At logic block 820, the modified scan insertion block beaks the loopback connections 440 inserted by the TR compiler 625 of the present invention. The loopback connections 440 or "routing" lines are broken because block 645 determines and routes the proper connections to form valid scan chains among the scannable sequential cells. The loopback connections 440 were temporarily installed during the HDL compile 625 steps to simulate the electrical characteristics of the actual scan chains for constraint optimization. At this time, they are removed. It is appreciated that blocks 815 and 820 can operate simultaneously.

At logic block 825 of FIG. 10, the present invention performs allocation of resources and routing to construct the scan chains between scannable sequential cells by linking non-violated scannable sequential cells. Also added during block 825 are other circuit resources required to construct the scan chains, such as signals lines for clock signals and scan enable signals, etc. The processing of block 825 required to determine the scan chains, allocate the proper resources to construct the chains, and route the input and output lines and gates accordingly is well known in the art. Any of a number of well known procedures can therefore be used consistent with the scope of the present invention at this stage to construct the scan chains. At block 825, the present invention will construct the scan chains differently, as is known in the art, depending if a full or partial scan is required.

Logic block 645 also determines if the added DFT implementations cause any potential circuit failures as a result of the application of test vectors. After scan insertion, additional logic may be needed to protect against potential problems that are encountered as a result of shifting test vector bits into memory cells that normally would never have these input values. For example, the application of the test vectors can cause bus shorts. In these cases, disabling logic elements are added by block 645 to prevent the potential short circuit.

At logic block 830 of FIG. 10, the present invention determines if a particular scan connection (e.g., between two consecutive scannable sequential cells) spans between modules. If so, then the present invention places an isolation gate between the output port of the upstream scannable cell and the downstream cell, this is shown in FIG. 7. This is referred to as hierarchical isolation. By providing the above isolation logic, the addition of the loopback connections 440 by the TR compiler 625 still accurately represents the electrical characteristics of a scan chain connection that spans across two modules.

At the completion of block 830, a number of components are added to the original circuit design that was input at block 810. These additions were not optimized to satisfy performance or design rule related constraints. Therefore, logic block 835 is provided by the present invention to provide a form of mapping optimization so that the overall design with the DFT circuitry can better meet performance and design rule constraints. These optimizations performed at block 835 are called "constraint driven" compile optimizations and are a proper subset of the optimizations performed by the TR compiler 625 of the present invention. However, as described below, block 835 provides certain optimizations applicable only to the DFT circuitry (e.g., the scannable sequential cells and the routing for the scan chains) in a tiered eIfort approach. In other instances, block 835 provides optimizations along critical points of the entire netlist design. The combination of the above techniques provides an eIfective optimization scheme for meeting design rule constraints and attempting to match performance constraints while not consuming a large amount of processing time.

Constraint driven compilation block 835 will be described in more detail with reference to FIG. 11A which shows processes to reach performance constraints and FIG. 11B which illustrates processes to reach design rule constraints (e.g., maximum fanout, maximum capacitance, maximum signal transition). Assuming the constraints are met by application of block 835, the output of block 835 is a scannable netlist of the circuit design.

With reference to FIG. 11A and FIG. 11B, the constraint driven compile block 835 of the present invention is described. Logic block 835 receives two selection controls as input (1) an eIfort selection (map.sub.-- effort) that can be low, medium or high and (2) an ignore design rules selection (ignore.sub.-- compile.sub.-- design.sub.-- rules) that can be true or false. Processing starts at block 850 where the present invention determines if a performance constraint (e.g., timing, etc.) exists and is violated. ff there are no performance constraints or none is violated, then block 835 flows to node "B." If a performance constraint violation appears, processing flows to logic block 852.

At logic block 852, the present invention performs a number of optimizations ("size design") along critical points only of the portion of the netlist that was introduced as a result of DFT processes. Within process 852, the entire design is not modified, only the portion or domain that comprises the DFT implementations, e.g., those portions that were added by insertion of the scannable sequential cells, the scan chains, and any other logic added as a result of the DFT implementations. The present invention is able to advantageously perform this type of size design process 852 because the present invention is knowledgeable about the DFT introduced circuits and can limit this application of the size design to these circuit portions. These optimizations are applied in accordance with design dependent heuristics that are described below. The optimizations performed on the critical points of the DFT introduced circuitry in block 852 include sizing, phasing, buffering, downsizing, isolation, off-loading, balancing, and splitting. These optimizations and their application heuristics are described in more detail in FIG. 12.

At block 852 of FIG. 11A, the cells that were added as a result of DFT are examined and a number of critical points and paths within the design are determined. These points generally contain negative slack values (e.g., they do not meet the timing constraints) and are ranked in a set. Therefore, according to one optimization employed at block 852, any drivers that belong to the DFT implementation at these points are increased in drive strength to increase the amount of slack at these points. By increasing the drive strength of these drivers at these points, the node has a better chance of meeting timing constraints. After a size increase in a particular driver is implemented, a new ranked set of critical points is generated. Other optimizations are then processed and the process is then repeated. It is appreciated that block 852 operates only on the DFT implementations.

At logic block 854 of FIG. 11A, the present invention checks if the map.sub.-- effort input indicated low effort. If so, then processing continues to node "B." If at the completion of block 852, the performance constraints are not violated and the map.sub.-- effort is medium or high, then processing also continues to node "B." If the map.sub.-- effort is medium or high and performance violations still exist, then processing flows to logic block 856. At logic block 856, the present invention performs the size design optimizations along critical points of the entire neflist and there is no domain restriction to the DFF implementations as with block 852. These optimizations are applied in accordance with design dependent heuristics that are described with reference to FIG. 12. The optimizations performed on the critical points of the circuit in block 856 include sizing, phasing, buffering, downsizing, isolation, off-loading, balancing, and splitting. They are described in more detail in FIG. 12. A substantial portion of the processing performed by constraint driven compile 835 is performed at block 856.

At logic block 858 of FIG. 11A, the present invention checks if the map.sub.-- effort input indicated medium effort. If so, then processing continues to node "B." If at the completion of block 856, the performance constraints are not violated and the map.sub.-- effort is high, then processing also continues to node "B." If the map.sub.-- effort is high and performance violations still exist, then processing flows to logic block 860. At logic block 860, sequential mapping is performed along critical points only for non-scan cells. At the identified critical points, surrounding combinational logic is combined with an adjacent non-scan sequential cell and both are replaced with a complex non-scan sequential cell that includes the functions of the surrounding combinational logic. The complex non-scan sequential cell is retrieved from the target technology library 743 as shown. The opposite can also occur wherein a complex non-scan sequential cell can be replaced with a simpler non-scan sequential cell and combinational logic. This process is similar to block 750 of FIG. 9. At block 860, functional equivalence is maintained during the mapping.

At logic block 862 of FIG. 11A, the present invention then reduces the size of circuits along points including non-critical points by performing local optimizations. Although this can decrease the drive strength of some circuits, it does not introduce performance constraint violations and has the advantage of bringing the design closer to the required area constraints. Once block 862 is complete, the present invention flows to logic block 864 to perform a particular local optimization where identified inverter pairs are replaced with single buffers reducing the number of extra or redundant inverters in the design. At the completion of block 864, the present invention then performs another size design optimization at critical points across the entire netlist. Logic block 866 is therefore analogous to block 856. Subsequently, processing flows to node "B" as indicated.

FIG. 11b illustrates the remainder of the processing performed by logic block 835 of the present invention. The flow enters through node "B" and at block 868 the present invention tests if the ignore.sub.-- compile.sub.-- design.sub.-- rules selection is true or false. If true, then processing flows to block 876 and logic blocks 870, 872, and 874 are skipped. Logic blocks 870, 872, and 874 represent optimizations the present invention can perform to ensure that the netlist satisfies the design rule constraints (e.g., fanout, signal transition, and node capacitance). Generally, by the addition of an appropriate number of elements, the present invention can fix the design rule violations. If ignore.sub.-- compile.sub.-- design.sub.-- rules selection is false, then processing flows to logic block 870 where the fanout number of each point is checked and violations of the fanout limitation in the design rules constraints are fixed. Violations are fixed at block 870 by adding an appropriate number of buffers so that each point along the design satisfies the fanout limitations.

At logic block 872 of FIG. 11b, the present invention checks the points in the design to determine if any signal transitions are too slow and violate signal transition constraints. The signal transitions can be too slow if the nodes are overloaded. To increase signal transition time, the present invention adds buffer elements to reduce the loading effects and/or increases the size the drivers at these nodes. At logic block 874, the present invention checks the points in the design to determine if any node has node capacitance that exceed the node capacitance constraints. If so, the present invention adds buffer elements to decrease the loading and therefore decrease the capacitance at any particular node having excessive node capacitance.

At logic block 876 of FIG. 11b, the present invention checks if map.sub.-- effort is low and if so, processing exits from block 835. If map.sub.-- effort is medium or high, then processing continues to logic block 878 where the performance constraints are examined. The addition of elements by blocks 870, 872 and 874 can cause certain performance constraints to be violated. If constraint violations are not present, then processing exits logic block 835. At this point, performance and design rule constraints have been met and the netlist is fully scannable.

If constraint violations are present at block 878 and this is the first pass through logic block 835, then processing flows from block 885 to node "A" which enters FIG. 11A at node 856. If constraint violations are present at block 878 and this is the second pass through block 835, then processing flows from logic block 885 and exits block 835. In the latter case, the constraint driven compile process 835 was unable to satisfy the performance constraints 610 of the design considering the DFT implementations.

By providing the processing flow shown in FIG. 11A and FIG. 11B, the present invention offers a three tiered effort driven constraint driven compile process. At low map.sub.-- effort, the least amount of CPU 101 processing time is consumed to size design only the circuitry added to the DFT implementations and design rules can be checked. At medium map.sub.-- effort, the above is done and the size design optimization is applied across critical points of the entire design. At high map.sub.-- effort, CPU 101 processing time is not a critical factor and the above is done as well as sequential mapping, size down, inverter reduction and another size design is performed. In addition, another size design optimization is performed. For medium and high map efforts, the above is performed through more than one pass if performance constraints are still not satisfied.

FIG. 12 illustrates in more detail the processing performed by the size design optimization block 856 (and block 866) of the present invention. It is appreciated that FIG. 12 represents a set of processes performed along critical points of the input circuit design domain in order to meet performance constraints. The size design optimization block 856 (of the constraint driven compile 835 block) is based on design dependent heuristics which determine processes are applied to the critical points of the input domain. The design domain can consist of only the DFT implemented circuitry (as for block 852) or can consist of the entire design (as for block 856).

Flow 856 of FIG. 12 enters at switch logic block 920 where design dependent heuristics determine which process to select from processes 902-916. At the completion of an optimization block (of 902-916), processing exists and can be repeated if critical points in the design remain. The blocks selected for subsequent processing depends on the design and history dependent heuristics and again is selected by switch logic 920. At block 902, the present invention increases the size and therefore the drive capacity of a driver or increases the receiving capability of a load. This is done to increase the response of a system to satisfy performance (e.g., timing) constraints. Block 902 is performed across critical points in the entire design. As discussed above, a critical point along the input netlist design consists of a design node wherein timing or other performance constraints are not met.

In logic block 904, the present invention moves a drive line from one output of a sequential cell to its inverse output. This is called phasing. FIG. 13A illustrates a sequential cell 1050 driving a load 1055 with the Q output. In phasing, the load 1055 is moved to the inverse output (/Q in this example) as shown in FIG. 13B. An inverter 1057 is required in this case. Block 904 is performed across critical points in the input design domain (e.g., the entire design or only the added DFT implementations). Phasing is particularly usefid in NMOS systems.

In logic block 906 of FIG. 12, buffering is performed where a number of different buffer configurations can be transformed. As shown in FIG. 14A, a single buffer 1061 can be transformed to a pair of inverters 1063 and 1065 as shown in FIG. 14B and vice-versa. A pair of buffers can also be replaced with a single buffer or vice-versa. At block 906, an inverter can be transformed into a buffered inverter or vice-versa. Or, lines coupled to a node in a branch network can be individually buffered. These options are applied across critical points in the input design domain to optimize to performance constraints. Options can be applied using a trial and error basis.

In logic block 908, the present invention performs downsizing where certain loads are determined that can be moved from one location to a logically equivalent location. After the load is removed, the associated driver is downsized to meet area constraints. FIG. 15A illustrates a sequential cell 1050 with an output, Q, driving a load 1055 and other lines 1071, 1072 driving other loads (not shown). Also shown is an other cell 1051. If is determined that Q of cell 1050 and Q of cell 1051 are logically equivalent, then the load 1055 can be moved from cell 1050 to cell 1051 as shown in FIG. 15B. In this example Q of cell 1050 is the critical point. The driver of cell 1050 is then downsized resulting in cell 1050'. By downsizing the driver, certain constraints can be more readily satisfied. Block 908 of the present invention is performed at critical points in the input design domain.

In logic block 910 of FIG. 12, the present invention performs load isolation. With reference to a critical point, all of the noncritical loads (NCLs) are buffered leaving the critical load (CL) directly coupled. In this way the NCLs are isolated from the critical point and path. FIG. 16A illustrates a critical point (Q) of cell 1050 coupled to two NCLs 1056, 1058 and one CL 1055. Isolation acts to buffer NCLs 1056 and 1058 with buffers 1073 and 1074 as shown in FIG. 16B. By isolating the critical point Q of cell 1055 from the NCLs, certain performance constraints can be more readily satisfied. Block 910 of the present invention is performed at critical points in the input design domain.

At logic block 912, the present invention performs offloading. In this process, the present invention determines NCLs of a particular critical point and offloads the NCLs to other equivalent drivers. The equivalent driver can have greater drive strength. Between many drivers, block 912 selects the one with the most slack. FIG. 17A illustrates two NCLs 1056, 1058 coupled to Q of cell 1050. Also a CL 1055 is coupled to Q. In offloading, the present invention offloads NCL 1058 to another driver. In this example, the equivalent driver is the /Q output of cell 1050 as shown in FIG. 17B. An inverter 1057 is added to maintain logical equivalence. Equally possible, both NCLs 1056, 1058 can be offloaded. Block 912 of the present invention is performed at critical points in the input design domain.

In logic block 914 of FIG. 12, the present invention performs balancing with respect to critical points of cells and within driver networks. With respect to a sequential cell, the present invention attempts to evenly balance loads across the Q and /Q outputs. This can require moving one load to another sequential output. With respect to a driver network, FIG. 18A illustrates a network having a common driver 1081 and secondary drivers 1083 and 1085. The loads 1085a-f are not balanced with more loads on driver 1085. The present invention moves a load from driver 1085 to driver 1083 to balance the network. The result is shown in FIG. 18B where the network is better balanced. By performing load balancing, certain performance constraints can be more readily satisfied. Block 914 of the present invention is performed at critical points in the input design domain.

In logic block 916, the present invention performs load splitting. In this process, drivers having critical loads are duplicated with the critical load applied to its own driver or a duplicate driver having increased drive strength. FIG. 19A illustrates a single driver 1081 with a critical load 1055 and other lines 1087 coupling with other NCLs (not shown). The present invention, as shown in FIG. 19B, duplicates driver 1081 and provides driver 1081' which is coupled to the critical load 1055. The remainder of the NCLs are coupled via lines 1087 to driver 1081. By splitting, certain performance constraints can be more readily satisfied. Block 916 of the present invention is performed at critical points in the input design domain.

By applying the above eight timing related optimizations to the input design domain, the present invention constraint driven compile provides effective measures to meet the performance constraints while requiring substantially reduced CPU 101 processing time over the prior art incremental compile step. Although a number of procedures can be used, an exemplary procedure is shown below to implement the processing of FIG. 12:

FIG. 20 illustrates that the present invention TR compiler 625 and the present invention modified scan insertion block 645 can be advantageously used to perform synthesis in a hierarchical fashion involving modules and submodules of an integrated circuit design. The present invention is extremely versatile in handling hierarchical processing because the modified scan insertion block 645 (1) accepts fully scanned, partially scanned or nonscanned netlist as input and also (2) operates efficiently in terms of processing time therefore chip level processing is possible without substantial delay.

Since the modified scan insertion block 645 accepts fully and partly scanned sequential cells, a number of different processes can output netlists to block 645. For those netlists that are fully scanned, block 645 performs no sequential translation or replacement but applies other procedures as described in FIG. 10 (such as loopback connection removal, scan chain routing, and optimization). For those netlists that are unscanned or partially scanned, block 645 performs the above tasks and also performs sequential replacement.

As shown in FIG. 20, a chip design can be represented by a number of modules 1001, 1003, 1005, and 1006. Alternatively, a single module can be represented by submodules 1001, 1003, 1005, and 1006. FIG. 20 illustrates an exemplary circuit synthesis using hierarchical approaches. Module A, in this example, is executed through the TR compiler 625 of the present invention that produces a fully scanned but unscannable netlist 1015. This netlist 1015 contains scannable sequential cells having loopback connections. A DFT DRC process can be run on netlist 1015 to mark violated cells. For reasons left up to designers, module 1001 is not processed on the module level by the modified scan insertion logic so the result is sent to block 1025.

Module 1003, like module 11301 is executed through the TR compiler 625 of the present invention which produces a fully scanned 1017 netlist that is nnscannable due to the loopback connections 440. A DFT DRC process can be run at this point to mark violated cells. The result is then processed by the modified scan insertion block 645 of the present invention. This process 645 will unscan violated cells. The result of block 645 is a scannable netlist 1021 meeting the defined constraints. This is forwarded to block 1025.

Module 1005 of FIG. 20 is processed through a compiler 225 that does not provide any scan replacement or DFT optimization. The result is a nonscanned netlist 1019 that does not contain scannable sequential cells or scan chains. This result 1019 is then fed through the modified scan insertion block 645 of the present invention which will provide sequential translation and construction of scan chains as well as optimization as shown in FIG. 10. The result 1023 is a scannable netlist having scan chains and scannable memory cells. This netlist 1023 can be mn through a DFF DRC to mark violated cells. This result is then forwarded to block 1025.

Lastly, module 1006 consists of a compiled design through a compiler without sequential replacement. Therefore, module 1006 contains unscanned cells that are technology dependent. This result is direcfiy forwarded to block 1025.

At the chip level, logic block 1025 builds a composition or chip level netlist 1027 combining the data received from all of the modules 1001, 1003, 1005, and 1006. The composition of modules step within block 1025 can be performed implicitly as a result of processing each module through their module level processing. Also at block 1025, certain binding or linking functions are performed on the design. Binding connects the ports of modules together as needed and also connects ports as needed at the chip level design.

Composition block 1025 also accepts as input a nonscannable netlist with scannable cells but no loop back connections 440. The netlist 1027 contains some fully scanned portions and some completely unscanned portions. The chip level netfist 1027 is then input to the modified scan insertion block 645 which produces a chip level scannable netfist 1029. Block 645 unscans violated cells. Block 645 of the present invention preserves the scan chains constructed within the design associated with netlists 1021 and 1023 while deriving original scan chains for the design associated with netlist 1015 and module 1006. These scan chains are then linked, as necessary, to provide chip level DFT implementations. It is appreciated that block 645 can also accept a chip level netlist including a portion comprising a nonscannable netlist with scannable cells but no loop back connections 440.

It is appreciated that the present invention modified scan insertion logic block 645 is advantageously suited for chip level scan insertion because it operates with substantially reduced processing time so that an entire chip level netlist can practically be processed. In other words, a time consuming full or incremental compile does not need to be performed on the chip level netlist 1027. Further, since block 645 accepts netlists with scanned and nonscanned sequential cells, it is suited for receiving input from a variety of different compiler options.

It is appreciated that designs 1001, 1003, 1005 and 1006 can also be subdesigns of a single module. In this case, block 1025 will process at the module level while the above tasks process at the submodule level.

The preferred embodiment of the present invention, a TR compiler is thus described that optimizes for DFT elements to meet constraints by inserting scannable memory cells in place of non-scan memory cells and by providing loopback connections for each scannable memory cell before optimizations are performed. While the present invention has been described in particular embodiments, it should be appreciated that the present invention should not be construed as limited by such embodiments, but rather construed according to the below claims.