Patent Grant
8010875
The present invention is related to the following commonly-owned, co-pending U.S. patent applications filed on even date herewith, the entire contents and disclosure of each of which is expressly incorporated by reference herein as if fully set forth herein. U.S. patent application Ser. No. (YOR920070268US1 (21189)), for “A SHARED PERFORMANCE MONITOR IN A MULTIPROCESSOR SYSTEM”; U.S. patent application Ser. No. (YOR920070293US1 (21233)), for “OPTIMIZED COLLECTIVES USING A DMA ON A PARALLEL COMPUTER”; U.S. patent application Ser. No. (YOR920070295US1 (21232)), for “DMA SHARED BYTE COUNTERS IN A PARALLEL COMPUTER”; U.S. patent application Ser. No. (YOR920070297US1 (21208)), for “MULTIPLE NODE REMOTE MESSAGING”; U.S. patent application Ser. No. (YOR920070298US1 (21209)), for “A METHOD AND APPARATUS OF PREFETCHING STREAMS OF VARYING PREFETCH DEPTH”; U.S. patent application Ser. No. (YOR920070299US1 (21212)), for “PROGRAMMABLE PARTITIONING FOR HIGH-PERFORMANCE COHERENCE DOMAINS IN A MULTIPROCESSOR SYSTEM”; U.S. patent application Ser. No. (YOR920070300US1 (21211)), for “METHOD AND APPARATUS FOR SINGLE-STEPPING COHERENCE EVENTS IN A MULTIPROCESSOR SYSTEM UNDER SOFTWARE CONTROL”; U.S. patent application Ser. No. (YOR920070301US1 (21210)), for “INSERTION OF COHERENCE EVENTS INTO A MULTIPROCESSOR COHERENCE PROTOCOL”; U.S. patent application Ser. No. (YOR920070302US1 (21216), for “METHOD AND APPARATUS TO DEBUG AN INTEGRATED CIRCUIT CHIP VIA SYNCHRONOUS CLOCK STOP AND SCAN”; U.S. patent application Ser. No. (YOR920070303US1 (21236)), for “DMA ENGINE FOR REPEATING COMMUNICATION PATTERNS”; U.S. patent application Ser. No. (YOR920070304US1 (21239)), for “METHOD AND APPARATUS FOR A CHOOSE-TWO MULTI-QUEUE ARBITER”; U.S. patent application Ser. No. (YOR920070305US1 (21238)), for “METHOD AND APPARATUS FOR EFFICIENTLY TRACKING QUEUE ENTRIES RELATIVE TO A TIMESTAMP”; U.S. patent application Ser. No. (YOR920070307US1 (21245)), for “BAD DATA PACKET CAPTURE DEVICE”; U.S. patent application Ser. No. (YOR920070321US1 (21256)), for “EXTENDED WRITE COMBINING USING A WRITE CONTINUATION HINT FLAG”; U.S. patent application Ser. No. (YOR920070322US1 (21255)), for “A SYSTEM AND METHOD FOR PROGRAMMABLE BANK SELECTION FOR BANKED MEMORY SUBSYSTEMS”; U.S. patent application Ser. No. (YOR920070323US1 (21246)), for “AN ULTRASCALABLE PETAFLOP PARALLEL SUPERCOMPUTER”; U.S. patent application Ser. No. (YOR920070324US1 (21264)), for “SDRAM DDR DATA EYE MONITOR METHOD AND APPARATUS”; U.S. patent application Ser. No. (YOR920070337US1 (21281)), for “A CONFIGURABLE MEMORY SYSTEM AND METHOD FOR PROVIDING ATOMIC COUNTING OPERATIONS IN A MEMORY DEVICE”; U.S. patent application Ser. No. (YOR920070339US1 (21292)), for “STATIC POWER REDUCTION FOR MIDPOINT-TERMINATED BUSSES”; U.S. patent application Ser. No. (YOR920070340US1 (21295)), for “COMBINED GROUP ECC PROTECTION AND SUBGROUP PARITY PROTECTION”; U.S. patent application Ser. No. (YOR920070355US1 (21299)), for “A MECHANISM TO SUPPORT GENERIC COLLECTIVE COMMUNICATION ACROSS A VARIETY OF PROGRAMMING MODELS”; U.S. patent application Ser. No. (YOR920070356US1 (21263)), for “MESSAGE PASSING WITH A LIMITED NUMBER OF DMA BYTE COUNTERS”; U.S. patent application Ser. No. (YOR920070357US1 (21312)), for “ASYNCRONOUS BROADCAST FOR ORDERED DELIVERY BETWEEN COMPUTE NODES IN A PARALLEL COMPUTING SYSTEM WHERE PACKET HEADER SPACE IS LIMITED”; U.S. patent application Ser. No. (YOR920070361US1 (21215)), for “HARDWARE PACKET PACING USING A DMA IN A PARALLEL COMPUTER”; and U.S. patent application Ser. No. (YOR920070371US1 (21335)), for “POWER THROTTLING OF COLLECTIONS OF COMPUTING ELEMENTS”.
1. Field of the Invention
This invention generally relates to computer systems, and more specifically, to detecting and correcting errors in data transmitted from a memory of a computer system.
2. Background Art
The small size of computer transistors and capacitors, combined with transient electrical and electromagnetic phenomena, cause occasional errors in stored information in computer memory systems. Therefore, even well designed and generally reliable memory systems are susceptible to memory device failures.
In an effort to minimize the effects of these memory device failures, various error-checking schemes have been developed to detect, and in some cases correct, errors in messages read from memory. The simplest error detection scheme is the parity bit. A parity bit is an extra bit included with a binary data message or data word to make the total number of 1's in the message either odd or even. For “even parity” systems, the parity bit is set to make the total number of 1's in the message even. For “odd parity” systems, the parity bit is set to make the total number of 1's in the message odd. For example, in a system utilizing odd parity, a message having two 1's would have its parity bit set to 1, thereby making the total number of 1's odd. Then, the message including the parity bit is transmitted and subsequently checked at the receiving end for errors. An error results if the parity of the data bits in the message does not correspond to the parity bit transmitted. As a result, single bit errors can be detected. However, since there is no way to detect which particular bit is in error, correction is not possible. Furthermore, if two or any even number of bits are in error, the parity will be correct and no error will be detected. Parity therefore is capable of detecting only odd numbers of errors and is not capable of correcting any bits determined to be in error.
Error correction codes (ECCs) have thus been developed to not only detect but also correct bits determined to be in error. ECCs utilize multiple parity check bits stored with the data message in memory. Each check bit is a parity bit for a group of bits in the data message. When the message is read from memory, the parity of each group, including the check bit, is evaluated. If the parity is correct for all of the groups, it signifies that no detectable error has occurred. If one or more of the newly generated parity values are incorrect, a unique pattern called syndrome results, which may be used to identify the bit in error. Upon detection of the particular bit in error, the error may be corrected by complementing the erroneous bit.
A widely used type of ECC utilized in error control in digital systems is based on the codes devised by R. W. Hamming, and thus take the name “Hamming codes”. One particular subclass of Hamming codes includes the single error correcting and double error detecting (SEC-DED) codes. As their name suggests, these codes may be utilized not only to correct any single bit error but also to detect double bit errors.
Another type of well-known ECC is the single symbol correction and double symbol detection (SSC-DSD) codes, which are used to correct single symbol errors and detect double symbol errors. In systems implementing these types of codes, the symbol represents a multiple bit package or chip. Hence, as the name implies, an SSC-DSD code in a system utilizing n bit symbols would be capable of correcting n bits in a single symbol and detecting errors occurring in double symbols.
As memory systems become larger, conventional single symbol error correct, double symbol error detect strategies, are no longer adequate. New strategies are required to deal with entire memory chip failures. Typically, several chips are used to hold user data with one or more additional chips used for check information and other required system data. If each cache line had two redundant chips, which could be used entirely for correction, then, for example, a conventional Reed-Solomon error correction strategy could be used. However frequently the system requires that a portion of the extra chips are allocated for system specific information, and the ECC has less than 2 full chips available for holding check information. Examples of such information are those that would reduce the power of the memory system, or improve the reliability of the memory system.
An object of this invention is to provide a strategy for detecting memory errors in computer systems.
Another object of the present invention is to provide a method and system for detecting and correcting memory chip errors when less than two full redundant memory chips can be used for error correcting code.
Another object of the invention is to provide a memory error correcting approach that, with high probability, can correct memory chip failure with a much-reduced amount of redundancy.
These and other objectives are attained with a method and system for detecting memory chip failure in a computer memory system. The memory system includes a first set of user data memory chips and a second set of system data memory chips. The method for detecting chip failure comprises the steps of accessing user data from the user data chips; and using error detection data from the system data chips testing the user data for errors. This testing step includes the steps of (i) generating a sequence of check symbols from the user data, (ii) grouping the user data into a sequence of data symbols, (iii) computing a sequence of syndromes using said sequence of data symbols and said sequence of check symbols, (iv) if all the syndromes are zero, identifying the user data as having no error, and (v) if one of the syndromes is non-zero, then (1) computing a set of discriminator expressions, and (2) using said discriminator expressions to determine whether a single symbol error has occurred or a double symbol error has occurred. In the preferred embodiment, if an error is detected, then the error is corrected, and less than two full system data chips are used for testing the user data and correcting the error.
Also, in the preferred embodiment of the invention, the step of computing the set of discriminator expressions includes the step of computing a set of discriminator expressions D0, D1 and D using the syndromes. More specifically, the sequence of syndromes includes first and second syndromes S0 and S1 respectively; and the step of using the discriminator expressions to determine whether a single symbol error or a double symbol error has occurred includes the step of testing to determine whether (i) D0=D1=D2=0, and (ii) both S0 and S1 are non-zero. If D0=D1=D2=0 and both S0 and S1 are non-zero, then a single symbol error has occurred, and the method comprises the further step of using S1 and S0 to identify the location and then correct the error. If each of D0, D1 and D2 is non-zero, then the method comprises the further step of computing a second set of discriminator expressions E2, E3 and E4. If E2=E3=E4=0 then a double syndrome error has occurred, and the method comprises the further step of using D0, D1 and D to identify the locations of the errors and compute the correct error values.
In addition, in one embodiment of the method of this invention, the syndromes may be used to determine if an entire one of the memory chips has failed. In this embodiment, the method comprises the further steps of identifying the number, N1, of syndromes in the sequence of syndromes; identifying the number, N2, of the data symbols in error when an entire one of the memory chips has failed; and determining a third number N3 by the equation: N1−N2=N3. N3 of the syndromes are modified to produce N3 modified syndromes, and these N3 modified syndromes are used to determine if an entire one of the memory chips has failed. In this embodiment, the N3 modified syndromes may also be used to identify the locations of said entire one of the memory chips that has failed and then compute the corrected contents of the memory chip. Once the location of the failed chip is known, then the N3 modified syndromes can be used to locate and correct an additional symbol error.
The basic approach of the preferred embodiment of the invention is to treat each potentially failing chip as a burst of errors and compute sets of modified syndromes for each possible chip location. If only one set of modified syndromes is zero, then we have located the failing chip. Since we want to minimize the performance impact on the memory system, preferably the ECC uses as much parallelism as possible to minimize latency.
Further benefits and advantages of the present invention will become apparent from a consideration of the following detailed description, given with reference to the accompanying drawings, which specify and show preferred embodiments of the invention.
Data bus 15, in one embodiment, is 160 bits wide but nevertheless may vary in width according to the requirements of the particular system while still receiving error protection under the ECC of the present invention. In this regard, the ECC of the present invention is specifically designed to meet the data requirements of a wide variety of servers including data buses having widths of, for example, 140, 142, 143, 144, 152, and 162 bits.
The data transmitted in computer system 10 is arranged into a data word having a size dependent on the particular data bus utilized by the system. Furthermore, the data word may be arranged into variably sized symbols. For instance, in one example, the data word comprises a plurality of six bit symbols.
With reference to
In our specific embodiment, we assume that the data to be protected in each line includes 128 bits of user data, two inversion bits (described below) and two address parity bits, which are not stored in the chips, but instead are computed from the memory address. Thus, there are 132 bits of data to be encoded per line for a total of 2*132=264 bits=44 six bit symbols. Assuming we use nine check symbols, we have a total codeword length of 44+9=53 symbols. Note that since our chips are 16 bits on each line, a chip failure produces six symbol errors. Using conventional error correction, we would need twelve check symbols to correct six symbol errors. The preferred embodiment of this invention can, with high probability, correct chip failures with a much-reduced amount of redundancy.
Check Symbols/Syndromes
We will first describe the encoding process which, given user data, will compute the nine check symbols. We will denote the 44 data symbols as d0, . . . , d43 and the nine check symbols as c0, . . . , c8. We then make the generator polynomial
We compute values for c0, . . . , c8 such that c0+c1x+ . . . +c8x8 is the remainder of dividing x9(d0+d1x+ . . . +d43x43) by g(x). This computation can be done by multiplying a fixed 9 by 44 matrix by the vector of 44 data symbols, whose columns correspond to the remainder by g(x) of the successive powers of x, from x9 through x52. We assume that the representation of our finite field is such that there exists an element u such that any six bit symbol can be split into two three bit fields, and multiplication by u maps the low order three bits to the high order three bits. In this case, we can operate separately on each line of user data, allowing us to cut the size of the encoder matrix in half. The final check symbols are computed by successively applying the half encoder matrix to each line of data, and adding the first application to u times the second application.
In the preferred implementation of the present invention, check bits are generated as data are transmitted from MSC 12 to memory 11. Upon generation, the check bits are stored, together with the transmitted data, in memory 11. As mentioned above, in the embodiment depicted in
To begin the decoding, we will first need to compute the nine syndromes. The received data r0, . . . , r52, where r0 through r8 correspond to the check symbols and r9 through r52 correspond to user data, is viewed as a polynomial
The syndromes S0, S1, . . . , S8 are computed as Sj=R (αj). This syndrome generation can also be done using a fixed 9 by 53 matrix. But since the lines of data arrive sequentially, we can also split this matrix into two and compute the contribution to the syndromes from the first line of data and combine the result from the second line of data times u.
If all the syndromes are zero, then the decoder presumes no error has occurred. Otherwise, in parallel, we check for various error patterns. One embodiment of this invention corrects a single symbol error, a double symbol error, or an entire chip failure (chip kill).
Single/Double Error Correct
The procedures for correcting single and double errors are illustrated in
If a double error occurred, then we must have that each of D0, D1 and D must be non-zero, and the error positions are determined by the two roots of the polynomial P (L)=D0L2+DL+D1. To find these solutions, at step 34, we compute T0=D0/D, T1=D1/D, T=T0T1, T2=D/D0. We now, at step 35, solve a “normalized” quadratic of the form M2+M+T, which can be directly solved with a small circuit, which produces one solution M (the other solution is M+1). The solutions to the original quadratic are now L1=MT2 and L2=L1+T2. Each of L1 and L2 can be represented as a power of α, and, as represented at step 36, the exponent determines the position of the error as in the single error case. The associated error values are computed as: e1=S0M+T0S1 and e2=e1+S0. Simultaneously we verify that two errors did actually occur by computing Ei=SiT1+Si+1+Si+2T0 for i=2, 3, and 4. For two errors to have occurred, all of the Ei must be equal to zero, as represented at step 37.
Chip Kill
The other class of errors which we want to handle involve memory chip failures (chip kill). In this particular embodiment of the invention, each memory chip delivers 16 bits of data per access. As our symbols have 3 bits from each of the two memory accesses, a 16-bit chip failure per line can produce six contiguous symbol errors. Our circuit operates in two modes. Before we have identified a known chip failure, we search for what we call a “soft chip kill”. This indicates a potential chip failure whose position is not yet known to us. Once the same chip has failed several times, it becomes a “hard chip kill”, which indicates that from then on we remember the location of the failing chip. After we have a hard chip kill, this invention will, in addition, allow the correction of a single symbol error event.
More specifically, for any of the ten chip positions, let the symbol L represent α raised to the exponent equal to the starting position of the chip, i.e. the first six bit symbol which has some bits in common with this chip. Now, form the polynomial
Relative to L, we can, at step 42, now form the three modified syndromes.
for j=0,1,2. Note that Ŝj is the inner product of the coefficients of P with seven syndromes starting in position j. So we can produce a modified syndrome matrix whose ten rows give the coefficients of P at all ten chip positions. If we multiply this matrix times the vector [S0, S1, S2, S3, S4, S5, S6] then we produce the ten modified syndromes Ŝ0 for all ten-chip positions. Using a copy of the same matrix, but applying it to the vector from S1 to S7, we would produce the ten values of Ŝ0 and finally if we again apply that matrix to the syndromes from S2 to S8, we produce the family of ten values for Ŝ2. To locate the failing chip, at step 44, we want a value of L such that the corresponding three modified syndromes Ŝ0, Ŝ1, Ŝ2 are all zero. If there is exactly one value of L with this property, then we have located the failing chip. The remaining task is to compute the six error values associated with the failing chip. This is done at step 46.
We begin by creating the inverse of the Vandermonde matrix V associated with the values 1, α, α2, . . . , α5. V is a six by six matrix such that Vi,j=aij for i,j=0, . . . ,5. Since V does not depend on L, its inverse V−1 is a constant matrix. The error values for the chip at location L are determined by [e0, e1, . . . , e5]=V−1[S0, S1/L, S2/L2, S3/L3, S4/L4, S5/L5], so we multiply the inverse Vandermonde times the first six scaled syndromes, where the ith syndrome is scaled by 1/Li. Note that we need to multiply by 5 scaling factors for each of the ten chip positions, so we can precompute these values and store them in a table. This completes the decoding in the case of “soft chip kill”.
After the same chip has failed several times, it becomes convenient to treat it as a permanently failed chip, which we call a “hard chip kill”.
Now we need to compute R2, R3, R4, R5, R6. The circuit to compute these powers can take advantage of the fact that squaring is a very cheap operation defined by a constant matrix. We can reuse one of the matrices we used for the modified syndrome applying it to the vector [1, R, R2, R3, R4, R5, R6]. The result will be 10 values of PL(R) and we extract the value corresponding to the value of L for the failed chip, i.e., take the ith value if chip i failed. Now the error value associated with position R is given by e=R=Ŝ0/PL(R). We then, at step 55, update the first five syndromes to remove the effect of the additional symbol error. Sj=Sj−eRj for j=0,1,2,3,4,5. We use these updated syndromes to correct the error value for the failed chip just as in the soft chip kill situation.
Invert Bits and Spare Bits
The ECC code above allows for 10 bits of system data in two transfers of data. These can be used to strengthen the memory system reliability, to reduce the power of the memory system, or any other such advantageous use. We chose to use the 10 bits (5 bits per transfer) as follows:
The invert bit is easily described. This technique is well known in the literature, however, we know of no instance where it has been included into an ECC field and thus protected. The bit is 0 if on the next transition of the bus, less than ½ the bits will change state. Otherwise, invert all bits including the “invert bit”. This bit is then checked when data is received. If set, invert all bits. T his is the simplest version of a class of codes known as “constrained switching” codes.
Here is a prescription for setting the invert bit. The data is being moved between an L3 cache built of EDRAM, which for purposes of this invention is similar to DRAM.
1) Receive data from L3 cache (32-byte bus).
2) Compare L3 data in 16 byte units. For example, compare the 2nd 16-byte transfer of data to the 1st 16-byte transfer of data. If more than 8 bytes switch, set inversion indicator for 2nd 16-byte chunk and store with data, use newly inverted data for the 2nd chunk to compare to 3rd 16-byte chunk. In our implementation, there is 1 invert bit per 8 bytes of data transferred.
3) When data is stored, generate ECC with inverted data and include inversion indicator in ECC matrix.
4) When data is fetched, verify data (i.e. ECC check) using the inverted data and the inversion indicator. Invert the data on its way back to the L3.
This flow is illustrated in
Here data stored in cache memory 100 is sent to logical unit 110, which analyzes each 8 Byte increment of data to be stored, by comparing it to the previous transfer. In parallel information is sent to external memory controller 105 informing it of an impending transfer so the external DRAM can be readied for write data. When 110 has completed its analysis for the first transfer of data and computed the invert bit value (and inverted all bits if necessary), the data is passed to store data buffer 120. In parallel, if bad bits were identified logical unit 115 replaces those bits with others by setting mux controls as is common practice. Also, address bus parity is computed by logic unit 130 and added to the store data buffer. When signal is received by 120 to move data out of the buffer, it is sent to ECC generation unit 125 and then written to memory. Here data statistics are kept to look for memory locations, which may need to be marked as bad, or replaced with redundant bits. If this had been a read from external memory, then data would have gone into logic unit 140 for possible ECC correction, and then passed to unit 135 for data inversion if the invert bits had been set, and finally stored in cache memory 100.
As will be readily apparent to those skilled in the art, the present invention or aspects of the invention can be realized in hardware, software, or a combination of hardware and software. Any kind of computer/server system(s)- or other apparatus adapted for carrying out the methods described herein—is suited. A typical combination of hardware and software could be a general-purpose computer system with a computer program that, when loaded and executed, carries out methods described herein. Alternatively, a specific use computer, containing specialized hardware for carrying out one or more of the functional tasks of the invention, could be utilized.
The present invention or aspects of the invention can also be embodied in a computer program product, which comprises all the respective features enabling the implementation of the methods described herein, and which—when loaded in a computer system—is able to carry out these methods. Computer program, software program, program, or software, in the present context mean any expression, in any language, code or notation, of a set of instructions intended to cause a system having an information processing capability to perform a particular function either directly or after either or both of the following: (a) conversion to another language, code or notation; and/or (b) reproduction in a different material form.
While it is apparent that the invention herein disclosed is well calculated to fulfill the objects stated above, it will be appreciated that numerous modifications and embodiments may be devised by those skilled in the art, and it is intended that the appended claims cover all such modifications and embodiments as fall within the true spirit and scope of the present invention.
This invention was made with Government support under Contract No.: B554331, awarded by Department of Energy. The Government has certain rights to this invention.
| Number | Name | Date | Kind |
|---|---|---|---|
| 4464753 | Chen | Aug 1984 | A |
| 4777595 | Strecker et al. | Oct 1988 | A |
| 4961193 | Debord et al. | Oct 1990 | A |
| 5063562 | Barzilai et al. | Nov 1991 | A |
| 5077737 | Leger et al. | Dec 1991 | A |
| 5142422 | Zook et al. | Aug 1992 | A |
| 5349587 | Nadeau-Dostie et al. | Sep 1994 | A |
| 5353412 | Douglas et al. | Oct 1994 | A |
| 5452432 | Macachor | Sep 1995 | A |
| 5459742 | Cassidy et al. | Oct 1995 | A |
| 5524220 | Verma et al. | Jun 1996 | A |
| 5634007 | Calta et al. | May 1997 | A |
| 5659710 | Sherman et al. | Aug 1997 | A |
| 5708779 | Graziano et al. | Jan 1998 | A |
| 5761464 | Hopkins | Jun 1998 | A |
| 5796735 | Miller et al. | Aug 1998 | A |
| 5809278 | Watanabe et al. | Sep 1998 | A |
| 5825748 | Barleu et al. | Oct 1998 | A |
| 5890211 | Sokolov et al. | Mar 1999 | A |
| 5917828 | Thompson | Jun 1999 | A |
| 6023732 | Moh et al. | Feb 2000 | A |
| 6061511 | Marantz et al. | May 2000 | A |
| 6072781 | Feeney et al. | Jun 2000 | A |
| 6122715 | Palanca et al. | Sep 2000 | A |
| 6185214 | Schwartz et al. | Feb 2001 | B1 |
| 6219300 | Tamaki | Apr 2001 | B1 |
| 6263397 | Wu et al. | Jul 2001 | B1 |
| 6295571 | Scardamalia et al. | Sep 2001 | B1 |
| 6311249 | Min et al. | Oct 2001 | B1 |
| 6324495 | Steinman | Nov 2001 | B1 |
| 6356106 | Greeff et al. | Mar 2002 | B1 |
| 6366984 | Carmean et al. | Apr 2002 | B1 |
| 6442162 | O'Neill et al. | Aug 2002 | B1 |
| 6466227 | Pfister et al. | Oct 2002 | B1 |
| 6564331 | Joshi | May 2003 | B1 |
| 6594234 | Chard et al. | Jul 2003 | B1 |
| 6598123 | Anderson et al. | Jul 2003 | B1 |
| 6601144 | Arimilli et al. | Jul 2003 | B1 |
| 6631447 | Morioka et al. | Oct 2003 | B1 |
| 6647428 | Bannai et al. | Nov 2003 | B1 |
| 6662305 | Salmon et al. | Dec 2003 | B1 |
| 6735174 | Hefty et al. | May 2004 | B1 |
| 6775693 | Adams | Aug 2004 | B1 |
| 6799232 | Wang | Sep 2004 | B1 |
| 6880028 | Kurth | Apr 2005 | B2 |
| 6889266 | Stadler | May 2005 | B1 |
| 6894978 | Hashimoto | May 2005 | B1 |
| 6954887 | Wang et al. | Oct 2005 | B2 |
| 6986026 | Roth et al. | Jan 2006 | B2 |
| 7007123 | Golla et al. | Feb 2006 | B2 |
| 7058826 | Fung | Jun 2006 | B2 |
| 7065594 | Ripy et al. | Jun 2006 | B2 |
| 7143219 | Chaudhari et al. | Nov 2006 | B1 |
| 7191373 | Wang et al. | Mar 2007 | B2 |
| 7239565 | Liu | Jul 2007 | B2 |
| 7280477 | Jeffries et al. | Oct 2007 | B2 |
| 7298746 | De La Iglesia et al. | Nov 2007 | B1 |
| 7363629 | Springer et al. | Apr 2008 | B2 |
| 7373420 | Lyon | May 2008 | B1 |
| 7401245 | Fischer et al. | Jul 2008 | B2 |
| 7454640 | Wong | Nov 2008 | B1 |
| 7454641 | Connor et al. | Nov 2008 | B2 |
| 7461236 | Wentzlaff | Dec 2008 | B1 |
| 7463529 | Matsubara | Dec 2008 | B2 |
| 7539845 | Wentzlaff et al. | May 2009 | B1 |
| 7613971 | Asaka | Nov 2009 | B2 |
| 7620791 | Wentzlaff et al. | Nov 2009 | B1 |
| 7698581 | Oh | Apr 2010 | B2 |
| 20010055323 | Rowett et al. | Dec 2001 | A1 |
| 20020078420 | Roth et al. | Jun 2002 | A1 |
| 20020087801 | Bogin et al. | Jul 2002 | A1 |
| 20020100020 | Hunter et al. | Jul 2002 | A1 |
| 20020129086 | Garcia-Luna-Aceves et al. | Sep 2002 | A1 |
| 20020138801 | Wang et al. | Sep 2002 | A1 |
| 20020156979 | Rodriguez | Oct 2002 | A1 |
| 20020184159 | Tadayon et al. | Dec 2002 | A1 |
| 20030007457 | Farrell et al. | Jan 2003 | A1 |
| 20030028749 | Ishikawa et al. | Feb 2003 | A1 |
| 20030050714 | Tymchenko | Mar 2003 | A1 |
| 20030050954 | Tayyar et al. | Mar 2003 | A1 |
| 20030074616 | Dorsey | Apr 2003 | A1 |
| 20030105799 | Khan et al. | Jun 2003 | A1 |
| 20030163649 | Kapur et al. | Aug 2003 | A1 |
| 20030177335 | Luick | Sep 2003 | A1 |
| 20030188053 | Tsai | Oct 2003 | A1 |
| 20030235202 | Van Der Zee et al. | Dec 2003 | A1 |
| 20040003184 | Safranek et al. | Jan 2004 | A1 |
| 20040019730 | Walker et al. | Jan 2004 | A1 |
| 20040024925 | Cypher et al. | Feb 2004 | A1 |
| 20040073780 | Roth et al. | Apr 2004 | A1 |
| 20040103218 | Blumrich et al. | May 2004 | A1 |
| 20040210694 | Shenderovich | Oct 2004 | A1 |
| 20040243739 | Spencer | Dec 2004 | A1 |
| 20050007986 | Malladi et al. | Jan 2005 | A1 |
| 20050053057 | Deneroff et al. | Mar 2005 | A1 |
| 20050076163 | Malalur | Apr 2005 | A1 |
| 20050160238 | Steely et al. | Jul 2005 | A1 |
| 20050216613 | Ganapathy et al. | Sep 2005 | A1 |
| 20050251613 | Kissell | Nov 2005 | A1 |
| 20050270886 | Takashima | Dec 2005 | A1 |
| 20050273564 | Lakshmanamurthy et al. | Dec 2005 | A1 |
| 20060050737 | Hsu | Mar 2006 | A1 |
| 20060080513 | Beukema et al. | Apr 2006 | A1 |
| 20060206635 | Alexander et al. | Sep 2006 | A1 |
| 20060248367 | Fischer et al. | Nov 2006 | A1 |
| 20070055832 | Beat | Mar 2007 | A1 |
| 20070133536 | Kim et al. | Jun 2007 | A1 |
| 20070168803 | Wang et al. | Jul 2007 | A1 |
| 20070174529 | Rodriguez et al. | Jul 2007 | A1 |
| 20070195774 | Sherman et al. | Aug 2007 | A1 |
| 20080147987 | Cantin et al. | Jun 2008 | A1 |
| Number | Date | Country | |
|---|---|---|---|
| 20090006899 A1 | Jan 2009 | US |
