flash memory is already a certain fact. In the past two or three years, the storage industry has promoted the 3D transformation of flash memory manufacturing processes, which has greatly expanded production capacity and affected the impact. The price of flash memory is also quickly reduced. Of course, for flash memory industry players, it may usher in an inevitable bloody battle, but for terminal manufacturers, it is good news. Especially SSD (solid-state drive) operators are expected to take advantage of this wave of decline in flash memory costs and quickly seize the mainstream storage position in the PC and consumer markets.
chart | Flash memory quotations have collapsed recently due to fierce competition, and SSD manufacturer fishermen benefited. Source: Chinese flash memory market
In the past, HDD (mechanical hard disk) has occupied the mainstream storage technology for decades, because the unit storage cost is relatively low and the performance is also higher than that of storage technologies of the same period. For example, floppy disks or optical disks are popular. In the past, with the rise of various types of video download demands in the Internet era, HDD sales have continued to rise. However, with the progress of SSD (solid-state hard disk) technology and the rise of network streaming, the application difficulties it faces are becoming increasingly difficult.
First of all, there is limited progress in transmission speed and random reading ability. Due to the inherent physical characteristics of mechanical HDD, it is very difficult to improve the reading and writing speed.
In addition, although from a traditional perspective, mechanical HDDs have high storage capacity and lowest unit storage cost, this is also the main reason why it was difficult to challenge its market position in the past by including optical storage or semiconductor storage. However, in terms of semiconductor storage, that is, solid-state HDD, due to the improvement of manufacturing process, the basic storage unit of solid-state HDD, that is, NAND flash memory, has rapidly reduced its manufacturing cost per unit of storage capacity, causing the storage cost per GB to decline rapidly, almost reaching a collapse, and has made traditional mechanical HDDs feel a lot of pressure.
Of course, in order to avoid being eliminated by the market, traditional HDD manufacturers such as Seagate have also actively strengthened the storage capacity and performance of HDD itself, Seagate has launched HAMR thermally assisted magnetic recording technology, and WD plans to import MAMR microwave assisted magnetic recording technology, which will increase the storage capacity of a single HDD by several orders of magnitude, hoping to continue to maintain the advantage of HDD's storage cost per GB. But the big trend has become, and in the future, HDD will be a direction that must be taken to the development of enterprise applications. Although
HDD can still maintain the advantage of unit storage cost, it is already a matter of time to be eliminated by the consumer market
Figure | Through the development of new generation magnetic storage technology, the storage cost gap per unit capacity between HDD and SSD should have the opportunity to maintain a magic number of 10 times in the future. Source: WD
Driven by new technologies, HDD still has considerable room for development in specific commercial storage applications, such as near-line and offline storage requirements in servers, and the video storage needs required by various types of monitoring industries, or the storage needs of movie workers for video editing, are all the HDDs that win by capacity.
Picture | Ultra-thin laptops are trendy, and HDDs basically have no room for survival in this type of product. Source: HP
However, in the consumer market, since the most mainstream HDD storage applications in the past, that is, the download and storage of online videos has become less popular, and video streaming services have replaced them, consumers' demand for storing videos on the local side has also decreased greatly. On the other hand, laptops are becoming increasingly thin, and traditional HDDs are basically unable to enter this lightweight computer organization.
On the other hand, for consumers, the operational response of computers depends on the choice of the storage system. Due to its huge advantages in continuous and random transmission performance, SSD is far better than HDD in system response capability. In the future, with the increase of bus bandwidth, the transmission performance of SSD may continue to extend upwards. However, HDD is basically out of reach for high-speed storage due to its innate mechanical structure. For consumers, the system response speed will also be the key to affecting their choice of storage media. Consumers who have used SSDs are basically unlikely to go back to endure the speed of HDD.
Finally, the flash memory market has entered an oversupply, and the surge in SSD capacity and the collapse in price are already happening. Although the capacity of SSDs in the enterprise market is still relatively small, it is impossible to expand to near-line and offline storage except online. However, in the consumer market that does not require massive storage, the living space of HDD will also be further squeezed out. In the future, HDDs may disappear in the consumer market.
3D manufacturing process drives the collapse of flash memory costs and accelerates the popularization of SSDs.
Currently, almost all flash memory used in SSDs has moved towards 3D technology. Indeed, 3D technology is actually much more expensive in terms of manufacturing costs compared to traditional planar technology, and its yield rate is worse than that of ordinary 2D technology. However, 3D technology has an advantage, that is, its ability to accommodate many layers can greatly increase capacity in the same chip layout.
currently stores related semiconductor parts, whether it is flash memory or run memory, and its pricing method is basically based on capacity units. In other words, if higher unit capacity is achieved under the same process basis, higher profits can be created.
. For every layer added to the 3D process, the capacity of the same area can be doubled. Judging from the mainstream 64 layers, the same chip area can reach 64 times the storage capacity of the original 2D process. If the 3D process is not used, the traditional 2D process is used and the unit area capacity is increased through process miniaturization. To reach 64 times, a process below 1nm is probably necessary, and it is basically unlikely to achieve it at a reasonable cost in the visible future.
The traditional 2D process is a process dominated by lithography. Nodes below 20nm usually require multiple multiple exposure and development steps. Of course, with the advancement of lithography equipment, single development and even node miniaturization have gradually become possible. In other words, the driving force for moving from one node to the next comes mainly from the improvement of lithography tools. When upgrading lithography tools, it is often possible to obtain improved tools with current tools for new ones, thus reducing conversion costs.
However, the improvement of this 2D process is actually extremely costly, and the manufacturing efficiency of lithography technologies of different generations is also different. Although it is possible to achieve improvements in unit storage costs from the advancement of 2D processes, the progress of 2D processes is gradually becoming more and more rapid. The capital cost and technical accumulation required for the development of processes below 10nm are no longer affordable by ordinary semiconductor chip manufacturers. Take TSMC as an example. The capital invested in the development of 7nm process will be nearly ten billion US dollars, and 5nm will even invest more than 25 billion US dollars. Although the process for DRAM/NAND is not much the same as the process used in logic chips, and the technical requirements are slightly lower, it is still considerable. In addition, the cost of technological innovation and investment in the future will only be slower and more expensive due to the limitations of semiconductor physical characteristics.
On the other hand, the minification of the 2D process will actually have a negative impact on the reliability of flash memory. Because the lattice structure of flash memory must be sufficiently thick to maintain electrical characteristics. If it is too thin, these structures may be penetrated by electrons during the read and write process, resulting in permanent data loss, which is also the so-called wear.
Since 2D cannot be used, I have to turn to 3D stacking, which is already a common manufacturing technology for flash memory used by SSDs or various memory cards.
diagram｜Comparing the cost structure of 16nm plane 128GB capacity TCL and 384GB capacity TLC manufactured with a 32-layer 3D process, although each wafer is nearly 2 times more expensive, the cost per GB is only about half. Source: Objective Analysis
The above figure shows that under the 3D process of 32 layers, the cost of the traditional 2D plane process increases by about 2 times compared to the traditional 2D plane process. In the case of 64 layers, according to the calculations made by Micron, the cost of the same wafer increases by about 5 times, which is also the trial calculation result after the additional yield. After the yield is improved, the available cutting rate of a single wafer will also increase, and the cost per unit capacity will further decrease.
In addition, 3D stacking usually uses backward processes, such as 32nm or 28nm processes, which is lower in cost, and secondly, the lattice of the flash memory can be maintained at a more reasonable thickness, which has obvious benefits for improving the service life of the flash memory.If calculated at the normal available yield, the flash memory cost starting from the 32 layer will gradually widen the gap with the flash memory of the plane process, and as the number of layers increases, the cost gap will become larger and larger. Of course, the yield rate after 32 layers is actually very difficult to improve. At that time, flash memory manufacturers such as Samsung spent nearly two years to improve the yield rate in the 64 layer process. Due to the low yield rate, in order to meet the customer's order needs, they had to make up for the insufficient yield through mass production, resulting in abnormal consumption of silicon wafers (the silicon wafers used in flash memory account for nearly 40% of all chip manufacturing applications), which is also one of the main reasons for the out-of-stock price increase of silicon wafers since the second half of 2016.
Since the second half of 2017, major flash memory manufacturers have basically achieved very high yields in the 64-layer process, and in 2018 they have also developed towards the 96-layer or even the 128-layer.
Due to the improvement of manufacturing processes and yields, flash memory costs have already shown a significant downward trend in the second half of 2018. Even major flash memory manufacturers, including Samsung, LG, Hynix, etc., are planning to control the speed of production expansion when inventory is full. However, the trend of multi-layering of 3D processes remains unchanged. Even if the expansion speed of equipment slows down, the growth of unit storage capacity cannot be stopped. Therefore, the trend of price decline in the next few years may be an unchanging trend. However, this is the norm in the production and inventory industry, and the industry is accustomed to it.
Figure | The cost difference between different processes, the cost per GB of 32 layers is 30% lower than that of the planar process, and the 64 layers are 30% lower than that of the 32 layers. Source: Micron
Figure | As the number of layers of 3D flash memory stack increases, the gap between storage cost per unit and plane process will continue to decline, and the 2D process may not be used in the future. Source: Micron
Due to the rapid maturity of 3D technology, flash memory costs have fallen in a crash. While the capacity of flash memory is constantly rising, prices have hit new lows. Although it is still a distance from the average capacity of HDD, under the limited demand for large-capacity data storage in general consumer categories, cost-effectiveness has gradually caught up with HDD. The inherent high-speed performance of SSD can greatly improve the user experience for consumers.
, and we can also see from the financial reports of several major HDD suppliers that HDD sales in the consumer market are also declining, and most sales are transferred to corporate or cloud customers. Instead, SSD sales targeting the consumer market are increasing.
Figure | The trend of storage cost per unit and performance comparison changes can also be seen that SSD has an absolute advantage in performance, and the cost is constantly approaching HDD, which accelerates the entire replacement process. Source: WD Blog
HDD manufacturers resist, mainly promote large-capacity technology and prioritize enterprise applications
. The two main HDD suppliers in the market, namely Seagate and WD, still embrace HDD technology in the market and look forward to maintaining their application life through technological changes. However, Seagate and WD are still quite different in product structure. Currently, nearly 90% of Seagate's revenue still comes from HDD. On the other hand, WD, HDD accounts for less than half of its revenue. Both of them regard SSD as an important strategic development goal in the future, but Seagate's flash memory needs to rely on external supply, while WD has mass-produced 64-layer 3D flash memory on a large scale, and the 96-layer product is also on the verge of being on the string.
, HDD's future market goals have almost always turned to enterprise applications as the main focus, supplemented by the consumer market, and capacity has become the biggest weapon.
, and Seagate and WD launch new technologies for the future HDD market as follows:
Seagate:
Heat-AssistedMagnetic Recording (HAMR)—Heat-assistedMagnetic Recording:
Figure | Seagate's HAMR technology has great capacity expansion space in the future, but is costly. Source: Seagate
General HDD Since the magnetic principle is used to record data, as the magnetic density increases, the difficulty of data stability will become increasingly difficult, which makes the current HDD magnetic density specifications slower.In order to increase the magnetic density while maintaining the stability of data, Seagate proposed HAMR (Heat-AssistedMagnetic Recording) many years ago. It uses lasers to accurately focus the area where data will be written, heat the medium, and avoid superparamagnetic effects of the magnetic medium. After the disk is heated to the Curiepoint by laser, the magnetic disc loses its magnetic properties and superparamagnetic effects. After the data is written, the disc will cool down quickly and the written data will stabilize. With such precise heating, HAMR can significantly increase the write density of the hard drive.
The ultimate storage density that can be achieved with this technology can reach up to 10TB/square inch. As for Seagate's official statement, HDD capacity specifications based on HAMR technology have a chance to reach more than 70TB in 2019.
However, the HAMR principle is easy to say, but it is not that simple to implement. HAMR requires brand new storage media, redesigned laser read and write heads, special NFT near-field optical sensors, and a large number of other new types of components. Therefore, the technology has not yet been able to achieve commercial mass production for so many years. However, Seagate seems to have made a technical breakthrough, and has recently announced that the first HAMR technology HDD is expected to be launched at the end of the year or early next year.
WD: Microwave-AssistedMagnetic Recording (MAMR)—Microwave-assisted magnetic record:
Figure｜WD's MAMR technology benefits are fewer key parts, lower manufacturing costs, but smaller capacity. Source: WD
Scientists at Hitachi Storage have invented another auxiliary magnetic recording technology, that is, microwave auxiliary magnetic recording, to increase write density from another technical perspective. Later, WD acquired Hitachi Storage in 2012, and naturally accepted this technology.
microwave auxiliary magnetic recording is to use the microwave field to act on magnetic moment , so as to increase the inversion speed of the magnetic moment and reduce the inversion field at the same time. When the magnetic moment precipitates under the action of the magnetic field, an resonance frequency will appear. The microwave auxiliary magnetic recording technology uses this resonance frequency just right to apply the auxiliary microwave magnetic field during the inversion and precipitation of the magnetic moment to promote the rapid reversal of the magnetic moment. Its core component is the Spin Torque Oscillator, through which the appropriate microwaves can be generated. After using MAMR technology, the theoretical density of the disc can reach up to 4TB/square inch.
Since the only key component of MAMR is the spin magnetic moment oscillator, the complexity and production cost are much lower than that of thermally assisted magnetic recording technology, its capacity amplification capability is slightly inferior to Seagate's HAMR technology. WD also plans to launch HDD based on MAMR technology in 2019, and its maximum capacity is expected to exceed 40TB. The new
technology is expected to prosper HDD, but it will turn to the enterprise market.
Although the storage cost per unit of HDD still has the opportunity to maintain a relatively low-end advantage after the new technology is imported, the storage capacity actually has its marginal benefits. After it exceeds a certain degree, it is no longer attractive to consumers. Instead, the storage transmission performance will become the key at this time, and the technical gap between HDD and SSD in this regard will probably only get bigger and bigger.
However, HDD is not useless. In the future, large-capacity HDDs can play the role of near-line and offline storage in servers. Since such applications do not require too fast speeds, they are expected to replace the functions of traditional tape drives and bring better and more reliable data backup capabilities to the operators.
