Dataset · FAST 2013 · FAST 2016

An example BCH code corrects raw bit error rates up to 1e-3, yet Google's field SSDs had median rates of 6e-10 to 3e-8, and raw rate did not predict uncorrectable errors

An example BCH code corrects raw bit error rates up to 1.0e-3 and soft LDPC up to 5.0e-3 (Cai et al., 2017; Zhao et al., FAST 2013 find about 3 times). Google's field median raw bit error rate was 5.8e-10 to over 3e-8, and 2 to 6 per 1,000 drive days had an uncorrectable error (Schroeder et al., FAST 2016).

Download CSV13 rows · 7 columns · CSV

File /data/datasets/nand-ecc-bch-vs-ldpc.csv · JSON metadata · All datasets · Human-readable note on hesela.com

Method

The figures are copied from the PDFs of Zhao, Zhao, Sun, Zhang, Zhang, and Zheng, FAST 2013, from the arXiv version of Cai, Ghose, Haratsch, Luo, and Mutlu, Proceedings of the IEEE 2017, and from Schroeder, Lagisetty, and Merchant, FAST 2016, not refit. Zhao et al. measure 25nm MLC NAND chips and drive a trace-based simulator of an SSD with a rate-8/9 LDPC code. Cai et al. is a survey whose BCH and LDPC strength curves are an example for similar codeword lengths and code rates. Schroeder et al. report raw bit error rates and uncorrectable errors from drives in Google data centers. The comparison of code limits with field error rates is this page's own and is made only to show orders of magnitude. Sources opened for this table: LDPC-in-SSD: Making Advanced Error Correction Codes Work Effectively in Solid State Drives, Zhao, Zhao, Sun, Zhang, Zhang, and Zheng, USENIX FAST 2013, pages 243 to 256 (USENIX PDF); Error Characterization, Mitigation, and Recovery in Flash-Memory-Based Solid-State Drives, Cai, Ghose, Haratsch, Luo, and Mutlu, Proceedings of the IEEE, September 2017 (arXiv 1706.08642, section 6); Flash Reliability in Production: The Expected and the Unexpected, Schroeder, Lagisetty, and Merchant, USENIX FAST 2016 (USENIX PDF, sections 3 and 4).

Limits

The code limits are example codes, not the codes inside any specific drive.
Field raw bit error rates are medians and percentiles of drive months; the tail inside a drive can be far higher.
Cai et al. cite Zhao et al., so the design sources are not fully independent.
The 3 times and 5 times gains use different codes and sensing levels.
Uncorrectable errors in the field are counted per drive day, not per bit.

Table

13 rows, the same rows as the CSV. Value low and value high are the printed range ends, or a single printed value in the high column. Units are in the unit column. Sources are the pages opened on 2026-10-11.
Measure	Value low	Value high	Unit	Scope	Note	Citation
Target uncorrectable bit error rate	not stated	0.0000000000000001	errors per bit read	Contemporary SSD makers, as cited by Cai et al.	Zhao et al. use 1e-15 as the decoding failure target.	cai-procieee17-pdf
Highest raw bit error rate an example BCH code corrects at that target	not stated	0.001	raw bit errors per bit	Code rate 0.935, one hard read	Example code chosen by the survey, not every drive's code.	cai-procieee17-pdf
Highest raw bit error rate an example soft LDPC code corrects at that target	not stated	0.005	raw bit errors per bit	Code rate 0.936, soft decoding with extra reads	Hard-decision LDPC alone is similar to BCH.	cai-procieee17-pdf
Extra reads of the example combined hard and soft LDPC scheme	not stated	0.28	percent more reads than BCH	If 0.01% of codewords fail and soft LDPC needs seven extra reads	Worst case, four codewords per flash page; the survey says this gives twice BCH's correction strength.	cai-procieee17-pdf
Raw bit error rate tolerance of soft LDPC relative to BCH	not stated	3	times	25nm MLC chip model, six extra sensing levels, 1e-15 target	Hard-decision LDPC tolerates only slightly more than BCH.	zhao-fast13-pdf
Redundancy of the LDPC code in the simulation	not stated	512	bytes per 4 KB of user data	Rate 8/9 code	The same overhead is used for the BCH comparison.	zhao-fast13-pdf
Worst-case read response time increase with soft decoding at 10,000 program/erase cycles	not stated	100	percent (lower bound)	Read-intensive traces, simulated SSD	Printed as over 100%.	zhao-fast13-pdf
Worst-case read response time increase with the authors' three techniques	not stated	20	percent (upper bound)	Same traces	Printed as below 20%.	zhao-fast13-pdf
Hard-decision LDPC decoding failure probability across the life of the flash	0.000000000000001	0.01	probability per read	Early life to late life	Soft decoding is only needed once failures become common.	zhao-fast13-pdf
Median raw bit error rate of first-generation drive models	0.00000000058	0.00000003	raw bit errors per bit read	Google data centers, several drive models	Printed as 5.8e-10 to more than 3e-8; MLC far higher than SLC.	schroeder-fast16-pdf
99th percentile raw bit error rate across models	0.000000022	0.000027	raw bit errors per bit read	Same drives; SLC-B lowest, MLC-D highest	The 99th percentile drive is usually at least 10 times the median of its model.	schroeder-fast16-pdf
Drive days with an uncorrectable error	2	6	per 1,000 drive days	Same drives	Correctable errors affect the majority of drive days.	schroeder-fast16-pdf
Drives with non-final read errors that retries can recover	not stated	2	percent of drives (upper bound)	Same drives	Printed as less than 2%; final read errors are about two orders of magnitude more common.	schroeder-fast16-pdf

Columns

measure (string): The quantity as the source names it.
value_low (number): Lower end of a printed range. Empty when the source prints a single value.
value_high (number): Single printed value, or the upper end of a range. For 'over' or 'below' it is the stated bound.
unit (string): Unit of the value columns: errors per bit, percent, times, bytes, probability, or per 1,000 drive days.
scope (string): Population and window the value applies to.
note (string): What the value is not, or the source wording behind it.
citation_id (string): Id of the opened source in the citations list.

License

Small derived table of figures printed in Zhao et al. (FAST 2013), Cai et al. (Proceedings of the IEEE 2017), and Schroeder et al. (FAST 2016), with credit. Not a Creative Commons license. The papers remain with their authors or publishers. This file is not a copy of the papers and not the data.