Table of Contents


Introduction

Scan-based Design is becoming an increasingly attractive Design for Test solution for ASIC designs as one considers increasing chip complexity and circuit density, increasing time to market pressures, as well as market pressures to provide the best hardware quality.

This document provides, in matrix form, a comparison between non-scan and scan solutions, and compares LSSD and non-LSSD scan implementations. It becomes clear that scan has significant advantages over non-scan:

It becomes additionally clear that LSSD-based design is an optimal scan solution, through:

Accordingly, LSSD is the Design-for-Test method employed by IBM ASIC products.

This document also investigates several methods of LSSD implementation for ASIC designs. The D-Mimic and Clock Splitter approaches are compared for the CMOS-4LP and CMOS-5L technologies, both for Standard Cell and Gate Array implementations. Through these comparisons, the following LSSD design recommendations become clear:

Comparison Matrix for Scan and

non-Scan

This matrix compares design structures using the following design parameters:

Test Coverage: Quality Measurement

Chip Area: A Hardware Overhead Measurement

Chip I/O: A Hardware Overhead Measurement

System Performance: A Hardware Overhead Measurement

Scan Performance: A Hardware Overhead Measurement

Isolation and Debugability: A Time-to-Market Measurement

Test Pattern Schedule: A Time-to-Market Measurement

Test Pattern Processing: A Cost Measurement

Test Equipment Cost: A Cost Measurement

Tester Time: A Cost Measurement

Enabled Benefits: Other Positive Factors enabled as a result of implementation of a given design structure

The following design structures are compared:

D-Mux Scan

D-Mimic LSSD

Leaf Splitter LSSD

See the Section "ASIC Scan Implementations" for a description of various Scan and LSSD approaches.

---------------------------------------------------------------------------------------------------------------------------
Factor                                                                  non-Scan         non-LSSD Scan    LSSD Scan          
---------------------------------------------------------------------------------------------------------------------------
Quality                                                                                                                      
Test Coverage                                                           80-90%           >99%          >99%            
Hardware Overhead                                                                                                            
Chip Area                                                               As necessary     9% (Note 1)      Leaf Splitter  7%  
(Note 1)                                                                to provide                        D-Mimic  15%       
                                                                        desired Func                      (Note 1)           
                                                                        tional Test                                          
                                                                        Coverage                                             
Chip I/O                                                                See above        Scan_Mode   1    ABC Clocks   3     
                                                                                         (Note 2)         (Note 2)           
System                                                                  See above        Mux (~300ps)     <100ps          
Performance                                                                              in Data Path                        
Issues                                                                                                                       
Scan                                                                    n/a              Possible short   None               
Performance                                                                              path race condi                     
Issues                                                                                   tions                               
Time to Market                                                                                                               
Isolation and                                                           Weeks            Days             Days               
Debugability                                                                                                                 
Test Pattern                                                            Months           Hours            Hours              
Schedule                                                                                                                     
Cost                                                                                                                         
Test Pattern                                                            Days/Weeks,      Hours            Hours              
Processing                                                              some designs                                         
(CPU)                                                                   require                                              
                                                                         hardware                                            
                                                                        accelerator                                          
Tester Cost                                                             Test equip       Enables          Enables            
                                                                        ment must        Reduced-Pin      Reduced-Pin        
                                                                        access All       Tester *         Tester *           
                                                                        Chip Pins                                            
Tester Time                                                             <10 seconds   ·  <10        ·  <10          
                                                                        for 80-90%       ·   seconds      ·   seconds for    
                                                                         coverage        for  80-         80-90%             
                                                                                         90%              coverage           
                                                                                         ·   coverage     ·  Tester time     
                                                                                         ·  Tester        increases          
                                                                                         time             for higher         
                                                                                         increases        coverage           
                                                                                         for  higher                         
                                                                                         coverage                            
*With Boundary Scan                                                                                                          
Note 1:  An estimate, based upon 20% of chip area used for                                                                   
registers.                                                                                                                   
Note 2:   Does not include SDI and SDO pins, which can be shared with                                                        
existing System I/O. There is also a capability to share the                                                                 
ABC clocks pins as well, reducing the pin overhead to a sin                                                                  
gle Scan_Mode input.                                                                                                         
---------------------------------------------------------------------------------------------------------------------------

TABLE 1: Comparison of Design Structures (Part1)

ASIC Scan Implementations

This section describes various ASIC implementations of scan-based design. Comparisons for effects on chip area, chip I/O, and performance are discussed. These comparisons are made against a non-scan register implementation using a D-FlipFlop, shown here:

FIGURE 1: D-FlipFlop for a Non-Scan Design Structure

This section will provide qualitative comparisons between scan implementations and DFF-based non-scan implementations. The following section, "Comparisons Between ASIC Scan Implementations", will quantify these comparisons in summary form for Standard Cell and Gate Array technologies.

D-Mimic LSSD Design

This design method is in use for some IBM ASICs. It provides a simple means for migrating a DFF-based design to an LSSD-based scan design. The following diagram shows how this is done:

FIGURE 2: D-Mimic LSSD Design

Comparison Characteristics

Register External View

Input---D-FF with SDI and Scan Clocks A,B,C

Output---D-FF with SDO

Register Logic SRL plus Clock Splitter for D-FF compatibility

Clock Tree Full Trees for A,B,C

System Data Delay Minimal

System Clock Effect None: Splitter is local to SRL

Scan Race Exposure None: LSSD

Added Clock Pins A,B,C

Other Considerations This method is extendable to the use of Asynchronous Set/Reset Latches

through the use of the D-Mimic with Asynchronous Set/Reset SRL.

Leaf Clock Splitter LSSD Design

The intent of this design is to reduce the cells per SRL overhead of the D-Mimic design by sharing the Clock Splitter over many SRLs, and thus significantly reduce the chip area overhead. However, to avoid possible performance limitations by placing a single Splitter at the root of the clock trees, a Clock Splitter is placed instead at each Leaf of the Clock Trees for the B, C, and edge clocks; this is the point in the clock powering tree where each buffer fans out to multiple SRLs, typically in the range of 10-20 SRLs. This approach is a good balance of design objectives that

  1. Spread the Clock Splitter area cost over many SRLs
  2. Minimize limitations on System Performance

FIGURE 3: Leaf Splitter LSSD Design

Comparison Characteristics

Register External View

Input---D-FF with SDI and Scan Clocks A,B,C

Output---D-FF with SDO

Register Logic SRL

Clock Tree Splitter at each Leaf of B,C,edge Powering Trees, full A Powering Tree

System Data Delay Minimal

System Clock Effects Minimal or None: Splitter placed after all Clock re-Powering

Scan Race Exposure None: LSSD

Added Clock Pin A,B,C

Comparisons Between ASIC Scan

Implementations

CMOS4LP Comparisons

--------------------------------------------------------------------------------------------------
Circuit                                                                Standard Cell  Gate Array    
--------------------------------------------------------------------------------------------------
SRL                                                                    CBLY=7         CBSS=12 ****  
D-Mimic SRL                                                            CBNI=8         CBIK=15       
D-FF                                                                   6*             CBVI=8        
D-FF w/Mux                                                             9*             11*           
Leaf Clock Splitter                                                    CBSF=12        CBH8=20       
*  Circuit is not currently offered. Number shown is a projection.                                  
****  Supporting pseudo book models not currently available. Contact                                
your Application Engineer for support in obtaining the option of                                    
CMOS4LP Gate Array with Leaf Splitter LSSD.                                                         
--------------------------------------------------------------------------------------------------
TABLE 2: CMOS4LP Cell Counts (Medium Power)

Note on D-FF's: The D-FF circuits described here are hypothetical, projected for the purpose of making chip area comparisons to the scan-based approaches. As such, although the following comparisons are based upon these D-FF's, such logic circuits do not actually exist in the design libraries.

Comparison of Scan Implementations

----------------------------------------------------------------------------------------------
Circuit                                 Scan      Total   System   System    Scan      Added    
                                        Cell      Area    Clock    Data      Data      Pins     
                                        Growth    Growth  Penalty  Penalty   Race      (beyond  
                                        (S =      *                          Exposure  SDI /    
                                        SRLs)                                          SDO)     
----------------------------------------------------------------------------------------------
Leaf                                    1.57 x S  5%      Minimal  No        No        3        
Splitter                                                                                        
D-                                      2.25xS    8%      No       No        No        3        
Mimic                                                                                           
D-Mux                                   3xS       10%     No       2-Input   Yes       1        
                                                                   Mux                          
* Based on DFF_area = 20% of chip_area                                                          
----------------------------------------------------------------------------------------------
TABLE 3: Comparison of Scan Implementations to
D-FF - CMOS4LP Standard Cell

D-Mux Scan-Based Design (non-LSSD comparison)

A D-FlipFlop can be adapted to Scan-Based design by applying a multiplexor to the Data Port. This is a non-LSSD approach to scan-based design, in that the edge clock is used as the scan clock, as well as for the system clock. This approach to scan has been used commonly in non-IBM applications. An exposure exists for race conditions due to the short path from SDO of one FlipFlop to SDI of the next FlipFlop in the scan chain, and must be corrected, or accounted for, in the system design. Additionally, the 2-way Mux in the data path results in an impact to System Performance.

The following diagram shows how this is done:

FIGURE 4: D-Mux Scan (non-LSSD Comparison)

Comparison Characteristics

Register External View

Input---D-FF with SDI and Scan_Enbl

Output---D-FF

Register Logic D-FF with Mux.

Clock Tree No additions

System Data Delay Mux Delay

System Clock Effects None

Scan Race Exposure--- Short Path Exposures exist for SDO to SDI path. Must be accounted for in the System Design.

Added Clock Pins None.

Other Considerations

Adds a Scan_Enbl pin

Scan must operate correctly at the edge clock frequency

This non-LSSD approach is not supported for IBM ASICs.

--------------------------------------------------------------------------------------------------
Circuit                                  Scan      Total     System   System    Scan      Added     
                                         Cell      Area      Clock     Data     Data      Pins      
                                         Growth    Growth *  Penalty  Penalty   Race      (beyond   
                                         (S=SRLs)                               Exposure  SDI/      
                                                                                          SDO)      
--------------------------------------------------------------------------------------------------
Leaf                                     4.97xS    12%       Minimal  No        No        3         
Splitter                                                                                            
D-                                       7.25xS    19%       No       No        No        3         
Mimic                                                                                               
D-Mux                                    3xS       8%        No       2-Input   Yes       1         
                                                                      Mux                           
*  Based on DFF_area = 20% of chip_area                                                             
--------------------------------------------------------------------------------------------------
TABLE 4: Comparison of Scan Implementations to
D-FF - CMOS4LP Gate Array

CMOS5L Comparisons

---------------------------------------------------------------------------------------------
Circuit                                                             Standard Cell  Gate Array  
---------------------------------------------------------------------------------------------
SRL                                                                 CHRG=9         CHHO=14     
D-Mimic SRL                                                         12 *           CHJ4=17     
D-FF                                                                7 *            9 *         
D-FF w/Mux                                                          12 *           14 *        
Leaf Clock Splitter                                                 CHQV=12        CHIJ=14     
*  Circuit is not currently offered. Number shown is a projection.                             
---------------------------------------------------------------------------------------------
TABLE 5: CMOS5L Cell Counts (Medium Power)

Note on D-FF:

The D-FF circuits described here are hypothetical, projected for the purpose of making chip area comparisons to the scan-based approaches. As such, although the following comparisons are based upon these D-FF's, such circuits do not actually exist in the design libraries.

------------------------------------------------------------------------------------------------------
Circuit                                   Scan Cell   Total    System   System    Scan Data   Added     
                                          Growth      Area     Clock    Data      Race        Pins      
                                          (S=SRLs)    Growth   Penalty  Penalty   Exposure    (beyond   
                                                      *                                       SDI/SDO)  
------------------------------------------------------------------------------------------------------
Leaf                                      2.57xS      7%       Minimal  No        No          3         
Splitter                                                                                                
D-Mimic                                   5.25xS      15%      No       No        No          3         
D-Mux                                     5xS         14%      No       2-Input   Yes:        1         
                                                                        Mux                             
*  Based on DFF_area = 20% of chip_area.                                                                
------------------------------------------------------------------------------------------------------
TABLE 6: Comparison of Scan Implementations to
D-FF - CMOS5L Standard Cell
----------------------------------------------------------------------------------------------------------
Circuit                                   Scan Cell   Total Area   System   System    Scan Data   Added     
                                          Growth      Growth *     Clock    Data      Race        Pins      
                                          (S=SRLs)                 Penalty  Penalty   Exposure    (beyond   
                                                                                                  SDI/      
                                                                                                  SDO)      
----------------------------------------------------------------------------------------------------------
Leaf                                      5.72xS      13%          Minimal  No        No          3         
Splitter                                                                                                    
D-Mimic                                   8.25xS      18%          No       No        No          3         
D-Mux                                     5xS         11%          No       2-Input   Yes:        1         
                                                                            Mux                             
*  Based on DFF_area = 20% of chip_area.                                                                    
----------------------------------------------------------------------------------------------------------
TABLE 7: Comparison of Scan Implementations to

D-FF - CMOS5L Gate Array

LSSD Model Support for Synopsys

The following 2 tables summarize IBM ASIC's LSSD Model Library support for Synopsys, in the CMOS4LP and CMOS5L libraries, for

  1. Edge-Triggered Flip-Flops
  2. Level-Sensitive "Transparent" Latches
Functional designs and characteristics of each of these circuits can be found in one of the following:

TABLE 8: IBM ASIC Supported Edge Triggered Models for Synopsys
----------------------------------------------
Block  Function            Technology  Library  
----------------------------------------------
CBIA   D-Latch (2 output)  CMOS4LP     GA       
CBTA   D-Latch             CMOS4LP     SC       
CBLY   D-Latch (dense)     CMOS4LP     SC       
CHHP   D-Latch             CMOS5L      GA       
----------------------------------------------
TABLE 9: IBM ASIC Supported Transparent Latch Models for Synopsys

References

  1. Level-Sensitive Scan Design: Concepts and Applications,Test Bench Version 1, Release 1 and all subsequent releases, " 1994 International Business Machines Corp.
  2. Using IBM LSSD Latches with Synopsys, by Lance Pickup, ASIC Products, IBM Microelectronics Division, " International Business Machines Corp.
  3. IEEE 1149.1 Boundary Scan in IBM CMOS-5L ASICs, by Steven F. Oakland, ASIC Products, IBM Microelectronics Division, " 1993, 1994 International Business Machines Corp.
  4. CMOS-4LP Logic Products Databook, IBM Microelecronics Division, " 1993 International Business Machines Corp.
  5. CMOS-5L Logic Products Databook, IBM Microelecronics Division, " 1993 International Business Machines Corp.