1. Application Overview
Personal computer processing power, SAN/NAS system complexity, transaction processing requirements and government-mandated record maintenance continue to push data storage capacities higher, both for individual users and the enterprise. Today’s storage modules, mostly magnetic disk drives due to cost-capacity curves, burn considerable power and generate considerable heat. And despite the need for immediate and highly random access, initial access latencies have diminished only slowly as capacity requirements continue to rise.
Long-awaited solid-state storage technologies such as NAND flash hold promise in density, power consumption, and access latency, but are either too expensive at relevant capacity points or have inherent long-term durability issues that the best wear-leveling techniques to date have failed to solve.
The SATAsphere and Bay Bridge SSDs are specially designed high performance SSD solutions. Leveraging unique, highly innovative, DuraClass™ technology, it leverages today’s densest Single-Level Cell (SLC) and Multi-Level Cell (MLC) NAND flash memory as well as ultra-dense next-generation MLC devices to enable Solid State Drives (SSDs) which greatly surpass magnetic hard disk drive performance in access latency and power consumed while guaranteeing long-term SSD reliability and converging rapidly on price parity with magnetic disks. The SATAsphere and Bay Bridge SSDs are the premiere SSD that enable MLC-based SSDs for enterprise applications, with both superior performance and assured product life. The SATAsphere and Bay Bridge SSDs require neither DRAM-based external circuitry nor overprovisioning of flash memory device count or capacity.
Additionally, the SATAsphere and Bay Bridge SSDs are the ideal solution for portable storage applications where power consumption, access speed, or storage drive ruggedness are important. Such applications as notebook computers, which combine a need for significant
volumes of data with a requirement for long-term data integrity, and which benefit greatly from extended battery operation time, high-speed access, and the inherent shock resistance of flash storage, are certain to be benefited from SATAsphere and Bay Bridge SSDs. Likewise, applications where storage solutions need not conform to traditional rotating disk drive form factors will benefit.
The SATAsphere and Bay Bridge SSDs enable solid-state storage to compete head-to-head in cost, capacity, and endurance with traditional rotating disks, while maintaining its clear advantage in ruggedness, form factor flexibility, access speed, and power consumed.
2. Theory of Operation
Utilizing unique, proprietary DuraClass™ technology, the SATAsphere and Bay Bridge SSDs accomplish the following:
- Greatly improves flash memory “endurance”
– Extends SSD life from months to years
– Enables warranties expected by enterprise and consumer markets
- Greatly improves flash memory data retention
- Provides orders of magnitude throughput improvement (as compared to magnetic disc drives)
- Maintains ultra-low latency advantages of flash memory
- Maintains ultra-low power consumption advantages of flash memory
– Solution requires no external DRAM
- Leverages declining cost-per-gigabyte advantages of MLC flash memory
– Solution requires no costly over-provisioning
- Leverages escalating density advantages of MLC flash memory
The DuraClass™ technology implements far superior, highly intelligent wear-leveling techniques to ensure that flash memory blocks are not worn out prematurely. Additionally, it intelligently minimizes re-writes of identical data, to maximize the effectiveness of the wearlevelling process. For any imaginable consumer or enterprise storage usage workload, the SATAsphere and Bay Bridge provide the first and most compelling workable solution.
The SATAsphere and Bay Bridge SSDs perform its usage enhancement techniques without need for costly, power-hungry external DRAM. It maintains all key advantages inherent in MLC NAND flash memory, while resolving key limitations.
2.1 Endurance vs. Retention
Flash memory technology offers significant promise toward the long-awaited solid-state data storage drive (SSD). But as all technologies, it has issues and imperfections. Its inherent limitations include:
- Endurance: Every write or erasure of a flash memory cell damages that cell very slightly. After enough writes or erasures (Program-Erase Cycles), the cell can no longer reliably store a data bit. flash memory blocks must be “retired” by an intelligent device as they become unreliable, making the overall capacity shrink. Depending on the geometry of the silicon process used in the manufacture, devices can be rated for a P-E cycle count of as little as several thousand writes or erasures. Given certain usage models, this can spell drive failure in as little as 30 days.
- Retention: When a cell is written or erased, the damage it suffers causes it to leak charge very slightly. Each write event increases the amount of leakage. As stated above, when the cell can no longer hold a charge, it is useless for data storage. But long before it reaches that condition, it is still leaking charge faster than when it was new. If the data it stores is not refreshed in time, the cell may still be a viable storage bit, but the data it held is gone. Retention can diminish from years to months and eventually to weeks.
It is clear that endurance and data retention are linked-in fact they are dependent variables of the same issue. Minimizing cell writes can thus not only extend flash memory usable life, it can extend data retention to the same degree.
DuraClass™ technology leverages highly effective proprietary write minimization techniques to avoid data storage duplication and reduce cell damage. Its intelligence is so effective that it can allow guarantees that an SSD will remain reliable throughout its warranted life, even with MLC flash. This efficiency intrinsically improves data retention to workable levels.
3. Read Disturb
Flash memory is primarily at risk from writes and erasures. However, reads also affect data longevity:
- Read-Induced Data Degradation: Excessive reads of flash memory cells induce inter-cell voltage shift, although the effect not as accelerated as write-induced cell damage. The degradation occurs in data stored in nearby cells, rather than in the cell being read. Read-induced data degradation is called “Read Disturb.”
The SATAsphere and Ba y Bridg e SSDs minimize Read Disturb issues such that expected data retention capability is assured throughout the warranted life of the SSD.
4. SATA Interface
The Serial ATA (SATA) interface is compliant with the SATA IO serial ATA specification, v2.6. The SATA interface connects the host computer to the SSD subsystem.
The SATA interface runs at a maximum speed of 3.0 Gbps. If the host computer is unable to negotiate a speed of 3.0 Gbps, the SATA interface automatically renegotiates to a speed of 1.5 Gbps.
5. Flash Memory Interface
For details please see Chapter 3.
5.1 Flash Memory Attributes
- MLC or SLC technology
- Flash device frequencies of 33.33 MHz, 40 MHz, and 50 MHz
- Popular densities including 32G bit die technology
- Popular timing parameters
- Single-die devices to 8-die stacks
- Dual-plane operation support
- Dual data bus device support
- Physical die/plane/block/page access
- Error correction applied by SATAsphere and Bay Bridge SSDs processor
- Write Protect support
5.2 DuraClass™ Technology
SATAsphere and Bay Bridge SSDs DuraClass™ technology includes the following unique features:
Write Operation Management
SATAsphere and Bay Bridge SSDs DuraClass™ technology embodies intelligent write management technology to make data location/relocation decisions which greatly increase the life of the SSD.
Wear levelling refers to the practice of equalizing the impact of write and erase operations over the larger pool of flash memory blocks. Industry standard wear levelling techniques focus on conventional schemes that attempt to equalize writes and erases across blocks.
While on the surface this appears to be a reasonable approach, it is clear that it assumes all blocks will “wear” equally when written or erased. This is far from the truth.
The SATAsphere and Bay Bridge SSDs take much more into account. It measures a variety of parameters to determine the actual wear of blocks during P-E cycles, to determine which blocks are impacted more by erasures and writes over time. That is, it determines actual cell wear, not simply assumed wear normalized to write/erase events.
The SATAsphere and Bay Bridge SSDs employ this information in its superior wear-levelling algorithm along with its ongoing record of writes and erasures, to ensure each block is impacted by P-E cycles no more than the average. The result is an SSD that is far more reliable across its full capacity and over a far greater length of time.
Write Operation Reduction
The SATAsphere and Bay Bridge SSDs use intelligent algorithms to minimize P-E cycles through agregation, virtualization, and difference processing. It is uniquely effective in reducing the wear and maintaining the reliability of the overall pool of flash memory blocks.
Read Operation Management
The SATAsphere and Bay Bridge SSDs overcome flash memory “Read Disturb” concerns by ensuring that data integrity is not impacted by multiple reads of the same flash memory address. It tracks reads and automatically and seamlessly recovers and refreshes data in proximity before that data is negatively impacted. Its superior throughput and latency performance, delivered over the life of the drive, is not diminished by this process.
Flash memory endurance and data retention are closely linked. By solving the endurance problem, the SATAsphere and Bay Bridge ensure that flash memory cells remain in a condition to hold data reliably throughout the warranted life of the SSD.
6. Data, Meta-Data, and Firmware Code Protection
The SATAsphere and Bay Bridge SSDs implement data protection throughout its data path. Protection techniques include:
6.1 DATA ECC
The SATAsphere and Bay Bridge SSDs can provide the following degrees of data error correction:
- Configuration options for flash memory devices providing 128 bytes of redundancy per 4K of data (normally this is SLC Flash)
– Standard Correction Power
> 16 bytes of redundancy applied to 512 bytes of data
> Up to seven 9-bit symbols (up to 63 bits if contiguous) correctable
– Extended Correction Power
> 34 bytes of redundancy applied to 494 data bytes
> Up to fifteen 9-bit symbols (up to 135 bits if contiguous) correctable
– Extreme Correction Power
> 43 bytes of redundancy applied to 485 data bytes
> Up to nineteen 9-bit symbols (up to 171 bits if contiguous) correctable
- Configuration options for flash memory devices providing 218 or more bytes of redundancy per 4K of data (normally this is MLC Flash)
– Standard Correction Power
> 27 bytes of redundancy applied to 512 bytes of data
> Up to twelve 9-bit symbols (up to 108 bits if contiguous) correctable
– Enhanced Correction Power
> 36 bytes of redundancy applied to 503 data bytes
> Up to sixteen 9-bit symbols (up to 144 bits if contiguous) correctable
– Extended Correction Power
> 45 bytes of redundancy applied to 494 data bytes
> Up to twenty 9-bit symbols (up to 180 bits if contiguous) correctable
– Extreme Correction Power
> 54 bytes of redundancy applied to 485 data bytes
> Up to twenty four 9-bit symbols (up to 216 bits if contiguous) correctable
6.2 RAISE™ Data Protection
Protection Against Catastrophic Flash Page/Block Failure
As part of its stand-out DuraClass™ technology, the SATAsphere and Bay Bridge SSDs implement proprietary R.A.I.S.E.™ (Redundant Array of Independent Silicon Elements) data protection, to overcome the probabilistic risk of page or block failure inherent in all flash memory technology.
Flash technology can exhibit a finite probability that a block or page will fail within the rated PE Cycle count lifetime of the flash device. While this probability may appear tolerable for a given application, note that it is for a particular flash die. For an SSD incorporating up to 128 flash die, the additive probability of this phenomenon can reveal measurable risk to the SSD over its multiyear lifetime.
DuraClass™ technology addresses this risk. In the event of a catastrophic failure of an entire flash page or flash block, RAISE™ off-line protection rebuilds the data in the failed page or block and relocates it elsewhere in the flash array. Performance during recovery is impacted, but after recovery is complete, the SATAsphere and Bay Bridge SSDs return to full performance and full functionality. The performance impact period is only the amount of time required to rebuild and relocate the page or block data, and to map out the problematic flash block.
In contrast to other SSD flash controllers, the SATAsphere and Bay Bridge SSDs with RAISE™ technology uniquely, reliably and seamlessly overcomes these catastrophic data loss risks with only temporary impact to throughput and latency and no impact to power consumption. In a RAIDed drive array application, the SATAsphere and Bay Bridge SSDs can auto-rebuild data locally, without passing the problem upstream to the system level and without incurring the associated significant system rebuild hit. The difference in impact between a standard approach and the RAISE™ approach is significant. Additionally, following recovery from a page failure or block failure, an SATAsphere and Bay Bridge SSDs are fully functional and fully reliable, whereas a page-failed or block-failed drive recovered by system RAID must be immediately replaced.
Protection Against Catastrophic Flash Die Failure
Flash technology also has a much more remote but still finite probability that an entire die will fail. RAISE™ technology can recover data from a failed die by rebuilding the data in off-line fashion for each read and delivering the data to the host. In the case of a failed flash die, the SF-1500 cannot return the drive to full performance functionality after rebuilding data, as relocating such a volume of data can be problematic. However, data is recovered. For non-RAIDed SSD applications, the benefit of RAISE™ technology in a failed die scenario is full data recovery (as opposed to data loss).
6.3 Data Path CRC Error Detection
SATAsphere and Bay Bridge CRC detection uses a 32-bit checksum (CRC32) to protect data along all internal data paths.
6.4 Data Protection in Internal Storage Structures
Phoenix employs various internal storage structures to assist its CPU and to manage data and meta-data. These storage structures may consist of:
- Internal RAMs
- Register Files
- Input/Output Buffers
- Lookup Tables
Parity-checking protects internal storage structures. Additionally, where storage structures reside between CRC generation and checking stations, CRC protection also applies.
6.5 Meta-Data Protection
Meta-Data is associated with user data within Phoenix. Meta-data is protected identically to user data. This includes parity specific to storage structures, CRC32 along meta-data paths, and ECC where the path falls within the ECC-correctable domain.
Representative SF-1500 performance targets are presented in Table 1 and Table 2.
|Table 1||Throughput Performance|
|Operation||Conditions 1||SATA Link Throughput|
|Sequential Reads||MLC or SLC Flash, 4-die stack, NCQ, 8K blocks and larger||to 240 Mbps|
|Random Reads||MLC or SLC Flash, 4-die stack, NCQ, 8K blocks and larger||to 240 Mbps|
|Sequential Writes||MLC or SLC Flash, 4-die stack, NCQ, 8K blocks and larger||to 240 Mbps|
|Random Writes||MLC or SLC Flash, 4-die stack, NCQ, 8K blocks and larger||to 240 Mbps|
|Table 2||IOPS Performance|
|Operation||Conditions 1||SATA Link IOPS|
|Sequential Reads||MLC or SLC Flash, 4-die stack, NCQ, 8K blocks and larger||to 30K IOPS|
|Random Reads||MLC or SLC Flash, 4-die stack, NCQ, 8K blocks and larger||to 30K IOPS|
|Sequential Writes||MLC or SLC Flash, 4-die stack, NCQ, 8K blocks and larger||to 30K IOPS|
|Random Writes||MLC or SLC Flash, 4-die stack, NCQ, 8K blocks and larger||to 30K IOPS|
Note 1: Conditions are examples only and depend on type of flash memory device.