Академический Документы
Профессиональный Документы
Культура Документы
Jin-Fu Li
Advanced Reliable S Ad d R li bl Systems (ARES) Lab. L b Department of Electrical Engineering National Central University Jhongli, Taiwan
Outline
Introduction Redundancy Organizations Built-In Redundancy Analysis Techniques Built-In Self-Repair Techniques Conclusions
Jin-Fu Li
Embedded MemoryQuality
During manufacture
Yield Exponential yield model Y = e AD , where A and D denote the area and defect density, respectively
After manufacture
Reliability
During use
Soft error rate
Jin-Fu Li
RAM
RAM
Memory Repair
Repair is one popular technique for memory yield improvement Memory repair consists of three basic steps
Test Redundancy analysis Repair delivery
Jin-Fu Li
Requirements: 1. 1 Memory tester 2. Laser repair equipment Disadvantages: 1. Time consuming 2. Expensive 2 E i
Jin-Fu Li
Function
Test Fault Location Redundancy allocation Swap Defective Cells
Jin-Fu Li
BIRA
RAM
BIST
Redundancy
Jin-Fu Li
RAM
Comparator
Jin-Fu Li
Redundancy Organizations
A memory array with local redundancies
10
Jin-Fu Li
Redundancy Organizations
A memory array with hybrid redundancies
Bank 1
11
Redundancy Organizations
A memory array with hybrid redundancies
Bank 1
Bank 2
Bank B k2
12
Redundancy Scheme
Three typical local redundancy schemes
Spare rows
Spare columns
Jin-Fu Li
Q1 c0c1c2c3
Q6 c0c1c2c3
Q7 c0c1c2c3
Q6
Q7
[ LogicVision]
14
Reconfiguration Scheme
32 columns Spare co S olumn Sense Amplifier p Decoder 10-bit 10 bit data Programming Module (Flash)
[M. Yarmaoka, et al., JSSC, 2002]
Advanced Reliable Systems (ARES) Lab., EE. NCU Jin-Fu Li 15
Sense Amplifier p
Firm reconfiguration
By B programming non-volatile memories to store i l til i t t repair information
Comparison
Advantages Soft
1. Multi-time repair 2. Low design overhead 1. Multi-time repair 2. 2 Short repair setup time
Disadvantages
1. Some latent defects cannot be repaired 2. Long repair setup time 2 L i t ti 1. High-voltage programming circuit is required q 1. One-time repair 2. Specific technology is required
Firm
Jin-Fu Li
17
BISR Strategies
Types of BISR
Off-line BISR On-line BISR
On-line O li BISR
19
Jin-Fu Li
64 Mb Memory Array
21 21
Spare p Memory M
16wx32b 16 32b
32
32
CAM
ROM
20
TPG
Comparator C
Jin-Fu Li
Main Memory
Spare Memory
Mux
RAM
Mux
An overflow bit identifies that there are more failing addresses than possible repair cells The programming of the faulty addresses is done during the memory BIST or from the f h f h fuse b d i memory box during setup
Advanced Reliable Systems (ARES) Lab., EE. NCU Jin-Fu Li 23
RAM
RAM Data Data Data Data Word Redundancy
Overflow
Advanced Reliable Systems (ARES) Lab., EE. NCU
TDO
Jin-Fu Li
Data out
[V. Schober, et. al, ITC01]
24
Furthermore, the application can be started immediately after the memory BIST passes i di t l ft th
This is called here soft repair
Jin-Fu Li
25
TDI
FA
Address
Data
TDO
Comparator
&
&
Read
Fail
Fail_address
Expected_data, DO
1 0
Fuse Boxes
The fuse box can be connected to a scan register to stream in and out data during test and redundancy g y configuration
Update Reset Fuse Box
FB
FB
FB
TDI
Scan FF
Scan FF
Scan FF
TDO
Fail
A[0]
Jin-Fu Li
28
BIST
FA
Fuse Address
FA
Fuse Address
FA
Address Register g
FA
Fuse Address
FA
Address Register g
FA
Jin-Fu Li
29
BIST
FA
Fuse Address
FA
Fuse Address
FA
Address Register
FA
Fuse Address
FA
Address Register
FA
TDO
Test finished?
No
Fail?
Yes
No
Fuse to be blown?
Yes
No
Jin-Fu Li
31
Redundancy Analysis
A repairable memory with 1D redundancy
Redundancy allocation is straightforward
4 5 6
Advanced Reliable Systems (ARES) Lab., EE. NCU Jin-Fu Li 32
1 0 1 1 0 1 1 2 2 1 0 0 1 1 0
Row Counters
Counters that indicate the number of faults that occur in a row, or a column
Apparently, the conventional software-based A l h i l f b d redundancy analysis algorithms are not adapted to be realized with hardware and be embedded into the li d i h h d db b dd d i h SOCs
Hardware overhead is too large
S1
S2
S3
S6
Finish ?
Yes
No
Each sub-analyzer analyzes in-coming row/column y y g addresses of faulty memory cells in parallel in a different repair strategy p gy Because CRESTA tries all the possible repair strategies of spare resources it guarantees finding a solution for a resources, repairable memory
37 Jin-Fu Li EE, National Central University
In an at-speed BISR design, a column repair vector (CRV) is used to store column failure information for solving this problem
CRV is a column repair vector of the same size as the word width
38 Jin-Fu Li EE, National Central University
At-Speed BIRA
Example of redundancy allocation
CCRR (Unrepairable)
RSV C C R R BIST Read Cycle 0 1 2 3 4 Fail_ Map 00000 10001 01110 01110 00000 Current Spare C1 C1 C2 R1 R1 Allocated Rows -----CRV 00000 10001 11111 11111 11111
0 0 1 2 3 4 1
4 1
2 3
2 3
2 3
CRRC
RSV C R R C BIST Read Cycle 0 1 2 3 4 39 Fail_ Map 00000 10001 01110 01110 00000 Current Spare C1 C1 R1 R2 C2 Allocated Rows --R1 R1R2 R1R2 CRV 00000 10001 10001 10001 10001 Jin-Fu Li EE, National Central University
In the BISRA, all C(m+n, m) analysis engines or just one engine can be implemented In one engine scheme, update the repair strategy if the current repair strategy fails and then re-run BIST and try the th next repair strategy t i t t
40 Jin-Fu Li EE, National Central University
BISRA Engine
BISRA Engine g
SRA
CAR
Restart
BISRA Engine
RSR
41
Jin-Fu Li
Repair Strategy Reconfiguration (RSR) block: updates the repair strategy and sends a restart signal to BIST controller restart
42
Jin-Fu Li
Jin-Fu Li
43
No
44
Must-repair phase M t i h
Row-must repair (column-must repair): a repair solution forced by a failure pattern with >SC (>SR) defective cells in a single row (column), where SC and SR denote the number of available spare columns and spare rows b f il bl l d
Final-repair phase
Heuristic algorithms are usually used, e.g., repair-most rule
Advanced Reliable Systems (ARES) Lab., EE. NCU Jin-Fu Li 45
SEG1
SR0 SR1 SR: Spare Row; SCG: Spare Column Group; SEG: Segment
46 Jin-Fu Li
SCG0
SCG1
Main Memory y
MAO
BIRA
POR
BIST
Spare Memory
BISR Flow
Power-on BISR procedure
Power On
Error information
BIRA
Continue C i
BIRA
Masked address
BIRA
Address Remapping
Address National Central EE,
48
Jin-Fu Li
University
Degraded Performance
Down-graded operation mode
If the spare rows are exhausted the memory is exhausted, operated at down-graded mode
The i Th size of the memory is reduced f th i d d
For example, assume that a memory with multiple blocks i bl k is used for buffering and the blocks are d f b ff i d h bl k chained by pointers
If some block is faulty and should be masked, then the pointers are updated to invalidate the block The system still works if a smaller buffer is allowed
49
Jin-Fu Li
Definitions
Definition
Subword
A subword is consecutive bits of a word Its length is the same as the group size
Example: a 32x16 RAM with 3-bit row address and 2-bit column address
BIRA Algorithm
Row-repair rules
To reduce the complexity, we use two row repair rules complexity row-repair
A row has multiple faulty subwords Multiple faulty subwords with the same column address and different row addresses
Examples:
subword
subword
Jin-Fu Li EE, National Central University
51
BIRA Procedure
BIRA procedure
Run BIST
Detects a fault Done
Stop
Met M t
Repair-Most Rules
Repair rate
The ratio of the number of repaired memories to the number of defective memories
A simulator has been implemented to estimate the repair rate of the proposed BISR scheme [Huang, et al.,
MTDT, 2002]
Industrial case:
SRAM size: 8Kx64 # of injected random faults: 1 10 1~10 # of memory samples: 534 RA algorithms: proposed and exhaustive search algorithms
53 Jin-Fu Li EE, National Central University
RR
18.37% 8.37% 73.10% 94.43% 99.26% 36.55% 36 55% 86.09% 99.26% 100% 72.17% 96.10% 99.81% 100% 72.36% 98.52% 100% 100% 85.90% 99.81% 100% 100%
RR (Best)
18.54% 8.5 % 86.14% 99.81% 100% 37.08% 37 08% 94.01% 100% 100% 55.06% 97.38% 100% 100% 71.91% 98.69% 100% 100% 85.77% 99.81% 100% 100%
EE, National Central University
0 4 8 12 0 4 8 12 0 4 8 12 0 4 8 12 0 4 8 12
0 1 2 3 0 1 2 3 0 1 2 3 0 1 2 3 0 1 2 3
69 16 1 1 46 3 0 0 18 3 0 0 8 0 0 0 6 0 0 0
45 9 3 0 18 8 0 0 7 2 0 0 5 0 0 0 1 0 0 0
32 7 2 0 13 0 0 0 7 0 0 0 1 0 0 0 0 0 0 0
Test Chip
Layout view of the repairable SRAM
Technology: 0.25um 0 25um SRAM area: 6.5 mm2 BISR area : 0.3 mm2 Spare area : 0 3 mm2 S 0.3 HOspare: 4.6% HObisr: 4.6% Repair rate: 100% (if #
random faults is no more than 10)
If a BISR circuit can be shared by multiple RAMs, then the area cost of the BISR circuit can drastically be reduced Shared BISR techniques h d h i
Reconfigurable BISR or IP-based BISR technique
Advanced Reliable Systems (ARES) Lab., EE. NCU Jin-Fu Li 56
BIST RSO
ReBISR
TDI LD
Advanced Reliable Systems (ARES) Lab., EE. NCU
Repair Process
Test & Repair
BIST BIRA Load Repair Signatures into the Fuse Group Pre-Fuse Testing Program Fuse
Normal Operation
Power-On Repair Signature Setup Normal Access
Jin-Fu Li
58
BIST RSO
ReBISR
TDI LD
Jin-Fu Li
59
Normal Mode
RAM 1 Wrapper RAM 2 Wrapper RAM N 1 N-1 Wrapper
BIST RSO
ReBISR
TDI LD
Jin-Fu Li
60
Syndrome n
Fail_h _
FSM
Syndrome Encoder
Bitmap
IO_Col
Address Ar+Ac
Address Masker
Remapping registers
TDI
A simulator was implemented to simulate the i l t i l t d t i l t th repair rate [R.-F. Huang, et. al, IEEE D&T, 2005 (accepted) Simulation setup
Simulated memory size: 4096x128 Simulated memory samples: 500 Poisson defect distribution is assumed Original yield is about 60% g y
Advanced Reliable Systems (ARES) Lab., EE. NCU Jin-Fu Li 62
Repair Rate
Case 1: 100% single-cell faults
Repair Rate (%) p ( )
100 90 80 70 60 50 40 29 8 29.8 29.8 30 20 10 0 (0,2) (0,4) (1,0) (1,2) (1,4) (2,0) (2,2) (2,4) (3,0) (3,2) (3,4) (4,0) (4,2) (4,4)
Jin-Fu Li
97.2 99.6 96.6 99.8 93.2 93 2 99.4 96.6 93.4 87.0 87.0
RCFA Opt.
12.6 12.6
(R,C) (R C)
63
Repair Rate
Case 2: 50% single-cell faults, 20% faulty rows, 20% faulty columns, and 10% column twin-bit faults
Repair Rate (%)
100 90 80 70 60 50 40 30 20 10 0
96.6 96 6 98.8 98 8 94.8 99.8 89.8 98.2 96.4 92.8 86.6 86.6
57.4 57 4 57.8
RCFA Opt.
(0,2)
(0,4)
(1,0)
(1,2)
(1,4)
(2,0)
(2,2)
(2,4)
(3,0)
(3,2)
(3,4)
(4,0)
(4,2)
(4,4)
(R,C)
64
Jin-Fu Li
ReBISR Implementations
FTC 0.13um standard cell library is used Three cases are simulated
Case 1 Core 0 Core 1 C Core 2 Core 3 64x2x8 128x4x16 128 4 16 256x8x32 512x16x64 Case 2 64x2x16 128x4x32 128 4 32 256x8x64 512x8x128 Case 3 64x2x32 128x2x64 128 2 64 256x4x128 512x4x256
Jin-Fu Li
65
Simulation Results
Delay d D l and area overhead h d
ReBIRA Parameter 512x16x64 512x8x128 512 8 128 512x4x256 Memory Area (um2) 1496258.4 1497561.6 1497561 6 1528848 ReBIRA Area (um2) 18766 20303 23255 Ratio (%) 1.25 1.36 1 36 1.52 Delay (ns) 2.5 2.5 25 2.5
BIRA time overhead w.r.t. a 14N March test with solid i h d M h i h lid data background
ReBIRA Parameter 512x16x64 512x8x128 512x4x256 Repair Rate (%) 83.6 82.3 83.8 ReBIRA Cycles 29952 30698 30404
Jin-Fu Li
MEMORY 2
BISR
MEMORY 0
C0 C1
SPAR ROW 0 RE
MEMORY 1
SPARE ROW 1 E E SPARE ROW 3
MEMORY 3
C2 C3
Jin-Fu Li
67
Infrastructure IP
What is Infrastructure IP
Unlike the functional IP cores used in SOCs the SOCs, infrastructure IP cores do not add to the main functionality of the chip. Rather, they are intended to chip Rather ensure the manufacturability of the SOC and to achieve lifetime reliability
Jin-Fu Li
68
Infrastructure IP STAR
STAR IIP
Mem. IW Mem. IW Mem. IW
Fuse Box
Mem. IW
STAR Processor 1
Mem. IW
[Y. Zorian, ITC02]
Jin-Fu Li
69
Infrastructure IP STAR
The infrastructure IP is comprised of a number of hardware components, including p , g
A STAR processor, a fuse box, and intelligent wrappers (IWs)
repaired
Soft Repair
repaired i d
Powered up unrepaired
Combinational Repair
Powered d P d down
Cumulative Repair
unrepaired
Jin-Fu Li
Infrastructure IP ProTaR
ProTaR [C.-D. Huang, J.-F. Li, and T.-W. Tseng, IEEE TVLSI, 2007
(accepted)]
Features
Parallel test and diagnosis Serial repair Support multiple redundancy analysis algorithms
Advanced Reliable Systems (ARES) Lab., EE. NCU Jin-Fu Li 72
ProTaR Controller
Wr rapper
ERR0 TGO0
RAM 0
ERRN-2
TGON-2
Shift_en
Wrappe er
RAM N-1
[C.-D. Huang, J.-F. Li, and T.-W. Tseng, IEEE TVLSI, 2007 (accepted)] Advanced Reliable Systems (ARES) Lab., EE. NCU Jin-Fu Li 73
Jin-Fu Li
74
R0 R1 R2 R3
RAR
RA algorithm Local repair-most (LRM) alg. Row first alg. Column first alg. alg
Advanced Reliable Systems (ARES) Lab., EE. NCU
Essential spare pivoting (ESP) alg. {FHFR ROW_FIRST, COL_FIRST} E ti l i ti l {FHFR, ROW FIRST COL FIRST}
ERR
Bitmap
FSM
[C.-D. Huang, J.-F. Li, and T.-W. Tseng, IEEE TVLSI, 2007 (accepted)] Jin-Fu Li 76
Addr_t DI_t
Bitmap
Addr DI DO TGOi ERRi
DO_S DO S
[C.-D. Huang, J.-F. Li, and T.-W. Tseng, IEEE TVLSI, 2007 (accepted)] Jin-Fu Li 77
The area of the four memories is 6798472um2 The area of all the redundancies is about 896060um2 The area of the IIP is only about 309893um2 Thus, th Th the area overhead of the IIP i only about h d f th is l b t 4.56%
Advanced Reliable Systems (ARES) Lab., EE. NCU Jin-Fu Li 79
[C.-D. Huang, J.-F. Li, and T.-W. Tseng, IEEE TVLSI, 2007 (accepted)] Jin-Fu Li 80
Conclusions
Embedded memories represent more and more area of system-on-chip (SOC) designs f t hi d i
The yield of memory cores dominates the yield of chips
Various BIRA techniques have been presented Different BISR techniques for memories in SOCs have b h been presented t d
Jin-Fu Li
81
References
1. 2. 3. 4. 5. T. Kawagoe, J. Ohtani, M. Niiro, T. Ooishi, M. Hamada, and H. Hidaka, A built-in self-repair analyzer (CRESTA) for embedded DRAMs, in Proc. Intl Test Conf. (ITC), 2000, pp. 567.574. V. Schober, S. Paul, and O. Picot, Memory built-in self-repair using redundant words, in Proc. Intl Test Conf. (ITC), Baltimore, Oct. 2001, pp. 995-1001. Y. Zorian, Embedded memory test & repair: Infrastructure IP for SOC yield, in Proc. Intl Test Conf. (ITC), Baltmore, Oct. 2002, pp. 340.349. C.-T. Huang, C.-F. Wu, J.-F. Li and C.-W. Wu,Built-in redundancy analysis for memory yield improvement, IEEE Trans. Reliability, vol. 52, no. 4, pp. 386-399, Dec. 2003. J.-F. Li, J.-C. Yeh, R.-F. Huang, C.-W. Wu, P.-Y. Tsai, A. Hsu, and E. Chow,A built-in self-repair scheme for semiconductor memories with 2-D redundancies, in Proc. IEEE Int. Test Conf. (ITC), (Charlotte), pp. 393-402, Sept. 2003. J.-F. Li, J.-C. Yeh, R.-F. Huang, and C.-W. Wu,A built-in self-repair design for RAMs with 2-D redundancies, IEEE Trans. Very Large Scale Integration Systems, vol.13, no.6, pp. 742-745, June, 2005. d d i l l S.-K. Lu, Y.-C. Tsai, C.-H. Hsu, K.-H. Wang, and C.-W. Wu, Efficient built-in redundancy analysis for embedded memories with 2-D redundancy, IEEE Trans. on VLSI Systems, vol. 14, no. 1, pp. 3442, Jan. 2006. T.-W. T T W Tseng, J.-F. Li C.-C. H A P K Chi and E Ch "A reconfigurable b il i self-repair J F Li, C C Hsu, A. Pao, K. Chiu, d E. Chen, fi bl built-in lf i scheme for multiple self-repairable RAMs in SOCs," in Proc. IEEE Int. Test Conf. (ITC), (Santa Clara), Paper 30.2, pp. 1-8, Oct. 2006. C.-D. Huang, J.-F. Li, and T.-W. Tseng,ProTaR: an infrastructure IP for repairing RAMs in SOCs, IEEE Trans. Very Large Scale Integration Systems vol 15 no.10, pp 1135 1143 Oct 2007. Trans Systems, vol.15, no 10 pp. 1135-1143, Oct. 2007 R.-F. Huang, J.-F. Li, J.-C. Yeh, and C.-W. Wu,RAISIN: a tool for evaluating redundancy analysis schemes in repairable embedded memories, IEEE Design and Test of Computers, vol.24, no.4, pp. 386396, July-August 2007.
6. 7.
8. 8
9. 10.
Jin-Fu Li
82