Ultra-Fast Storage: Dive Into SSDs
The difference between SSDs and HDDs
As we know, the consumer hard drives are dominated by HDDs (mechanical hard drives) and SSDs (solid-state hard drives). HDDs are available in 2.5-inch (for laptops) and 3.5-inch (for desktops) physical sizes, as well as SSHD (hybrid hard drives) with built-in high-capacity flash memory. There are five connector types that SSDs use to interface with a computer, including SATA, mSATA, M.2, U.2, and PCIe, which apply to different PCs respectively.
SSD (solid-state drives)
Capacity/price ratio: 1GB/10 cents
Sequential read speed: extremely fast
Random read/write speed: extremely fast
P/E cycles: depending on the flash unit structure
Working noise: zero noise
Data recovery: Extremely difficult
SSDs, without the complex mechanical structure of HDD, are electronic chip-based storage devices composed of NAND flash memory chip, master control, and cache (optional). The more flash memory chips, the larger the storage space. Factors affecting SSDs performance include unit structure of NAND flash memory, master control chip model, cache capacity, etc. The main advantages of SSD are fast, lightweight, no fear of vibration, and high reliability.
HDD (hard disk drive)
Capacity/price ratio: 1GB/ 4 to 6 cents
Sequential read speed: Slow
Random read/write speed: extremely slow
P/E cycles: unlimited
Working noise: the higher the rotating speed, the louder the noise will be
Shockproof: extremely poor
Data recovery: easy
The internal structure of HDD is relatively complex, consisting of a disk disc, magnetic head, spindle, and driveshaft, and other mechanical structures. The principle of HDD is similar to that of the old phonograph, except that each disk disc in HDD corresponds to two magnetic heads up and down, and the spindle speed is much higher than that of the phonograph. Therefore, HDD can be damaged by vibration and bumping, and consumes relatively high power when reading/writing data.
SATA, mSATA, and M.2 are common form factors in SSD families, but the mSATA SSD has been historically obsolete, while U.2 and PCIe have not been.
M.2 SSD (C)
Maximum speed: 32Gbps
Specification/length: 42mm, 60mm, 80mm
Interface: PCI-E×2, PCI-E×4, SATA
Introduction: M.2 is the mainstream high-speed SSD form factor at present. Under the adoption of the PCI-E×4 channel and NVMe protocol, its maximum read speed can easily break through 3000MB/s.
SATA SSD (A)
Maximum speed: 6Gbps
Specification/length: 2.5 inches
Introduction: 2.5-inch hard disk is common on laptop. SATA3.0 is also the most popular hard disk interface. So most laptops support SATA SSD. However, its bottleneck is the maximum read speed is only about 550MB/s.
mSATA SSD (B)
Maximum speed: 6Gbps
Specification/length: 50mm, 30mm
Introduction: The mSATA SSD can be seen as a mini SSD. Its interface is used to be exclusive for early ultrabooks and mini-computers, however, it has been replaced by the more mini and high-speed M.2 SSD.
Maximum speed: 32Gbps
Specification/length: 2.5 inches
The U.2 SSD, while still in the form of a 2.5-inch hard drive, can deliver up to 32Gbps and has a new physical interface. U.2 SSDs are expensive and not popular enough to be ignored by ordinary users.
Maximum speed: 32Gbps
Specification/Length: over 170mm (length)
Interface: PCI-E×2, PCI-E×4
PCIe SSDs can be plugged directly into the PCI slots of PC motherboards, and their PCB motherboards can be composed of multiple M.2 SSDs to form a RAID array, which is the most powerful hard disk form factor. However, the high cost also doomed it to be the choice of a few advanced users.
Structural decomposition of SSD
No matter what kind of SSDs they are, the most core components are NAND flash memory, DDR cache, and master control chip. The reason why there is a big difference in price and performance between SSDs with the same shape and capacity is that they are affected by the specifications of these three chips. So let’s take a look at these core units of SSDs.
Chip name: DDR cache (A) (top left chip)
Purpose: Low power DDR4 SDRAM for caching data. Theoretically, increasing the cache chip and cache capacity can prolong the SSD life and gain some performance. If you want to de-cache, you need a new master and algorithmic support.
Chip Name: Master control chip (B) (second left the chip in the above picture)
Purpose: The hub to connect DDR cache, NAND flash memory, and PC. The speed of an SSD largely depends on the model of the main control chip. Moreover, we can improve SSDs’ performance and extend the life through a firmware update.
Chip Name: NAND flash memory (C) (third and fourth chips from left above)
Purpose: NAND flash memory is a better storage device than traditional hard drives. It uses non-volatile storage technology, which is a storage device that can hold data after power failure. The storage units for SSDs, flash drives, phones, tablets, and other devices are all NAND flash memory.
The evolution of NAND Flash
Since its inception, SSDs has evolved toward greater performance, higher capacity, and lower prices. Achieving these three goals starts with NAND flash, including the introduction of more advanced production processes, multi-tier units that can store more data, and a shift from 2D tiling to 3D stacking structures.
Like CPUs we are familiar with, NAND flash chips are cut from silicon wafers and their electrical properties are affected by the manufacturing process. As CPUs switch to more advanced manufacturing processes (such as upgrading from 14nm to 10nm), more transistors can be integrated into smaller chips and the leakage rate between grids can be improved, resulting in higher frequency (performance) and lower power consumption. The same is true of NAND flash chips, which more advanced manufacturing process can achieve greater capacity with the same chip area.
Silicon wafer processes have been slow to upgrade to keep pace with market demand for SSD capacity. Therefore, starting from the number of unit layers of NAND flash memory is an effective means to alleviate this contradiction. NAND flash memory is made up of several units in bits that record the storage state (0,1) by having the charge turned on or off.
Name: SLC (single-level cell)
Introduction: a typical storage unit, whether DRAM, SRAM, or ROM, can only record 1bit of data. In NAND flash memory, these 1-bit units are called “single-tier units” or “SLC”.
P/E cycles: 100,000
Pros: Most powerful read/write performance, most read/write cycles, low read/write error rate, better reliability
Cons: Small capacity, high cost
Status and buy advice: Suitable for servers or applications that require a lot of reading and writing daily. The consumer market has largely disappeared.
Name: MLC (multi-level Cell)
Introduction: MLC is a multi-layer unit, which features a 2bit/ unit based on the 1bit/ unit of SLC. It can record four states of 00, 01, 10, and 11 so that more data can be stored in the same chip area, which improves the storage capacity.
P/E cycles: 10,000
Pros: Powerful read/write performance, lifespan and stability than TLC has a huge advantage
Cons: Not as durable and reliable as SLC
Status and buy advice: For heavy PC users and gamers. And only some of the older, discontinued products still use MLC flash
Name: TLC (three-level unit)
Introduction: TLC flash memory can hold 3 bits/unit and can record eight states of 000, 001, 010, 011, 100, 101, 110, and 111, further increasing the storage density, but its read/write cycle is also greatly shortened.
P/E cycles: 500 to 1000
Pros: capacity greatly increased, cost significantly reduced
Cons: Significant performance and lifespan degradation
Status and buy advice: suitable for ordinary users in the non-professional field. TLC flash memory has been unified SSD arena.
Name: QLC (quad-level cells)
Description: QLC 4-layer unit is the cheapest specification in flash memory production, 4 bits/unit can record 16 states of 0000,0001…1111, and so on, compared with TLC, the storage density increased by 33%, but its performance and lifespan were further reduced.
P/E cycles: 150 to 1000
Pros: Capacity can be further enhanced and cost can be further reduced
Cons: Relative worst performance, lifespan is roughly equal or slightly worse than TLC.
Status and buying advice: For users seeking capacity vs. price. And there will be more and more cheap SSDs based on QLC flash in 2019
Extended reading: Why is performance getting worse from SLC to QLC
The cost of upgrading the NAND flash unit level is increased design difficulty and reduced performance. It is assumed that the flash working voltage is 2V, and a bit of SLC flash has two states, each of which can be evenly distributed with 1V voltage. The four states of MLC can evenly distribute 0.5V voltage; When it comes to QLC, each state can only allocate 0.125V, and more states and smaller voltages will lead to more difficult control and serious interference problems, which will affect the performance, reliability, and stability of flash memory.
Although NAND flash memory can increase capacity and reduce cost through process improvement, more advanced processes will also lead to a thinner oxide layer of NAND and thus affect reliability. Meanwhile, NAND flash goes from SLC to MLC to TLC to QLC, each time drawing questions and abuse. To solve these problems, 3D NAND flash stack and Silicon Perforation (TSV) packaging technologies were born.
The so-called 3D NAND (Samsung also calls its technology V-NAND) is a kind of “build-in-a-building” stacking technology that adds more capacity to a single chip’s area. Samsung has already mass-produced 96-layer 3D NAND SSD (970 EVO Plus SSD) in 2018, Toshiba and Kingston have released 96-layer 3D NAND SSD (BG4 and A2000), and in early 2019 Toshiba even announced that 128-layer 3D NAND particles have been developed.
In the past, NAND flash memory was mainly based on 2D packaging. To gain more capacity, an SSD needs to embed more NAND flash chips on the PCB motherboard at the same time. The capacity of 128GB may needs 10 chips, which takes up the whole space of the PCB motherboard.
There are many benefits of 3D NAND technology, for example, 64 layers of 3D NAND can achieve 1GB capacity on a single particle, significantly reduce cost and power consumption, and even extend the theoretical P/E cycles of QLC flash memory from 150 to 1000!
Advanced applications of flash memory
There is no end to consumer’s demand for storage performance. When SSDs are large enough and cheap enough, how can we achieve greater performance? Intel, Samsung, and Western Digital have their answers.
Optane memory is based on the 3D XPoint storage technology developed by Intel and Micron. It has the performance close to memory and the characteristics of data storage that can be saved by power loss. Several key indicators, such as delay, P/E cycles, and speed, are also much better than traditional flash chips.
Optane memory (system cache, using the M.2 interface)
Optane hard drive (ultra high speed SSD, using U.2 or PCIe interface)
Samsung Z – NAND
Z-Nand is a 64-layer stacked MLC flash memory that runs in SLC mode. Where it is advanced is in encapsulation. It’s for the highest level of enterprise storage for supercomputing, AI analysis, and so on, compared to where Intel’s Optane technology is advanced in storage media.
Ultra-high-speed SSD (using PCI interface)
LLF NAND of Western Digital
The LLF (Low Latency Flash) memory chip will have a cost per GB between DRAM memory and traditional 3D Flash and uses SLC or MLC particles with a latency of microseconds.
Suitable for working environments requiring low latency, high performance, and long lifespan.
The importance of main control chip
For SSDs, the main control chip plays the role of “brain”, which is essentially an embedded processor designed based on ARM or RISC architecture. Its performance is affected by the number of cores, frequency, the number of flash channels supported, and the interface bus. The main control chip of a 2.5-inch SSD is unlikely to make much of a difference in performance, but the main control chip of M.2 SSD can vary depending on the flash particle, flash channel, and interface bus.
The role of main control chip:
1. Coordinated work among multiple flash memory chips
2. Responsible for data transfer between flash memory and external interface
3. Complete internal instructions through various firmware algorithms, such as error checking and correction, wear balancing, bad block mapping, Read Disturb, cache control, GC garbage collection, Trim instruction, and data encryption, etc.
Currently, the new-generation main control chip supports 64 or higher levels of 3D NAND flash, and also introduces the optimization of QLC flash. As more PCs (including laptops) come with the M.2 slot as standard, PCIe NVME SSDs become the main battleground for SSDs. In general, the performance of SSDs will be affected by the level of flash particles (TLC or QLC) and the number of flash particles when the main control chip is the same, and the more flash particles there are in the same capacity, the better the write speed will be.
Distinguish between good and bad NAND flash memory
The price difference between a first-tier brand and a copycat brand for the same volume of SSDs can be as much as two times. In addition to the differences in brand and main control chip, the source of NAND flash particles is more critical, only the original chip and the ones bought from famous brands can ensure 100% stability, ink die are a bit less, as for those based on fake die, we have to stay away from any cheaper SSDs.
|Graded nand flash chips||Introduction||Identification||Quality|
|Original chip||the original chip made by Intel, Micron, Western Digital (SanDisk), Toshiba, Samsung and Hynix, etc.||have a laser marking from the original chip supplier||outstanding|
|Good die||buy the wafer from the original chip supplier and then cutting the wafer by themselves||have a laser marking from SSD manufacturers||excellent|
|Ink die||there are a few die/district which is not stable during storing data but does not affect the normal use||is usually printed SSD vendor identification or only product code||general|
|Fake die||non-passed chips from the finally test-passed and are worse in storing data||without any identification or the fake manufacturer identification||poor|
As consumers, it’s hard to know what kind of flash memory particles are embedded in an SSD before you pay for it, and it’s hard to tell from the identification on the chip surface. Therefore, the author hopes that we can recognize the fact that SSD is “you get what you pay for”, and select products from top brands first, even if the performance is poor, there is no quality worry.
Understanding the parameters of SSD
SDD has numerous brands and models. If you want to select the most suitable products from them, you should start with their official parameters.
|Nowadays, SSDs are mainly in two form factors: traditional 2.5-inch and M.2. The 2.5-inch SSD is limited by the SATA3.0 interface, and its sequential read and write speeds are mostly 500MB to 550MB, and the random read and write speeds are difficult to exceed 10KIOPS. The advantage of M.2 SSD is that you can choose higher-speed PCI-E3.0 x 4 channels and NVMe protocol (PC device compatibility is required), to achieve more extreme performance.|
|Available capacity||500GB||500GB||SSD will generally set OP reserved space, used for various optimization tasks of the main control chip, and coupled with the difference in unit conversion will have different capacities of 480GB, 500GB, and 512GB|
|Sequential read speed||560 MB/S||3,400 MB/S||can be understood as the maximum theoretical speed when SSD reads large files|
|Sequential write speed||530 MB/s||2,500 MB/s||can be understood as the maximum theoretical speed of SSD when writing large files.|
|Random read speed||95000IOPS||410000IOPS||IOPS is the number of reads and writes per second of SSD, but according to the difference between the test and the use environment, it does not guarantee the correctness and reference of the value. This value will be affected by the ratio of reads and writes, the factors such as the number of threads, the depth of access queues, the size of data blocks, system settings, drivers and operating system background programs, and other factors.|
|random write speed is as high as||84000IOPS||330000IOPS||ditto|
|TBW||200||300||The data amount that an SSD can write during lifespan. The higher the TBW value, the longer the lifespan of the SSD|
If you’re looking at SSD performance, prioritize the higher sequential/random read/write speeds between SSDs of the same capacity and price. SSDs are generally guaranteed for 3 or 5 years. Considering that SSDs are difficult to retrieve internal data once they are scrapped, we need to pay more attention to the parameter of “TBW”, through which we can calculate the maximum amount of data that SSDs can write every day during the warranty period.
The maximum amount of data written per day = (SSD’s TBW * 1000)/(365 * years)
Suppose that a 500GB SSD has a 3-year warranty and 200TBW durability. The above data can be substituted into the formula:
The maximum amount of data written per day = 200 * 1000/(365 * 3) = 182GB
That is unless you can write more than 182GB of data into an SSD every day, you should theoretically be able to scrap a 500GB SSD during a three-year warranty period. SSDs of top-tier brands does not need to worry about their lifespan under normal conditions. Taking the Samsung 860 QVO using QLC flash memory for example, its 1TB, 2TB, and 4TB versions have 360TBW, 360TBW, and 1440TBW respectively, which cannot be used up during the 3-year warranty period.
The compatibility of SSD
By the time you read here, I believe you have a preliminary understanding of SSD. The next step is to buy one. Considering the two common form factors of SSDs, we need to treat them separately.
As for a 2.5-inch SSD, as long as you have a SATA slot on your PC, you can choose it. For laptops, you’ll need a 2.5-inch hard drive slot or an optical drive extension (which requires a special stand) to install. Since the SSDs are all based on the SATA3.0 bus, their performance does not differ much, so warranty period and TBW are often more meaningful than speed.
The buy advices
1. Priority of flash memory chip: MLC > TLC > QLC
2. Priority of 3D NAND : 96 layers > 64 layers > 32 layers
3. Priority of performance: random performance > sequential performance
4. The longer the warranty period and the larger TBW, the better.
Before purchasing M.2 SSDs, compatibility issues should be considered first, namely whether Socket 2 (B key) or Socket 3 (M key) is adopted for M.2 SSDs on the PC motherboard, and then compatible M.2 SSDs should be selected according to this slot and budget. If the motherboard slot is “B key”, unfortunately, it can only use the SATA bus’s M.2 SSD. If the socket is the “M Key” standard, then it can be compatible with SATA or PCIe bus and NVMe protocol’s M.2 SSD.
The interfaces on M. 2 SSDs can be divided into three standards: “B key”, “M Key” and “B&M Key”. Among them, the SSD of the “B key” interface has long been discontinued, and all the M.2 SSDs you can buy are “M key” or “B&M Key”. Among them, the M.2 SSDs of the “M Key” interface adopts PCI-E×2 and PCI-E×4 buses and only supports the same standard motherboard slot. The M.2 SSDs of the “B&M Key” interface, on the other hand, are almost all SATA buses and can be installed on any M.2 slot on the PC motherboard.
The buy advices
1. The priority of flash memory chip: MLC > TLC
2. The priority of 3D NAND: 96 layers > 64 layers > 32 layers
3. The priority of Bus: PCI-e ×4 > PCI-e ×2 > SATA (incompatible case)
4. The priority of performance: random performance > sequential performance
5. The longer the warranty period and the larger TBW, the better.