digital storage projections for 2019, part 2 - energy storage devices

This is the second of the three of us.
2019 part of the digital storage projection blog series.
This section introduces 2018 trends and future developments in solid state storage, memory and basic computer architecture.
In particular, we will look for the development of fast storage interfaces, especially NVMe and structures built with NVMe (NVMe-oF).
This article will also explore emerging non-
Loss-prone memory and its in-computing storage and in-
Memory calculation.
Finally, we will explore the trend of driving high-performance solid-state memory in IoT and data center environments.
We discussed the pricing of NAND flash memory in the first part of the 2019 projection series.
Start in the middle
The shortage of 2018 NAND flash memory ended and NAND prices began to decline.
The price decline will continue to reach 2019, possibly until some point in 2020.
The decline in prices is due to a decline in demand in some markets (e. g. phones)
3D with major NAND vendors-
NAND factory to achieve full production.
The 2018 3D NAND accounts for most of the market shipments.
At the same time, the industry is focused on proposing its 96-
Layer technology, including four layers and three layers of unit technology.
There was an announcement in 2019 for more than 100 layers of NAND, although Changjiang memory from China said they would focus on 128-layer NAND flash memory in 2020. 96-
Layer NAND flash memory should start mass manufacturing in the second half of 2019, from 64-
NAND flash layer.
However, several announcements in 2018 suggest that by stacking layers of "strings" together, the 3D NAND layer may exceed 500 at some point in the next decade.
The following figure shows a view of the gradual addition of this layer.
Please note that in order to avoid the process time getting out of control, it takes a lot of effort to reduce the thickness of the layer as the stack gets higher and higher.
This increase in the NAND flash layer should help to continue the overall price decline of NAND flash memory, thus continuing to use NAND-built storage devices.
Four-level unit technology is especially suitable for putting extra pressure on hard drives, although the endurance of solid-state drives built using this technology may not match the hard drive.
Still, lower-priced SSDs using this technology may help the industry further penetrate markets such as laptops and desktop computers.
NVMe interface developed in the past few years, using non-volatile solid-
Remember that state storage provides significant performance improvements compared to the SAS and SATA interfaces originally developed for hard drives.
Various forms of NVMe SSDs are being used more and more in client, consumer and enterprise applications.
SSDs with NVMe NAND flash memory are particularly popular as primary storage in the data center (the cloud).
In 2018, enterprise NVMe SSD shipments exceeded SATA and SATASSDs.
Decline in NAND prices and growth in cloud storage and cloud storage
Deep learning based on applications, especially for big data analytics and the internet of things, will drive huge demand for data center high-performance NVMe SSDs in 2019 and beyond.
Structure-based NVMe is developed to provide NVMe protocol support through various network structures.
While this has been used on Infiniband and fiber channel for more than a year, official TCP support for NVMe was released in 2018.
Other NVMe features, such as IO Terminism, are expected to be launched in 2019, which will provide more options for NVMe SSDs (
For example, different parts of the SSD can have different uses and different endurance features). NVMe-
The emergence of products using Ethernet connectivity began in 2018, and these products will surge in 2019.
Providing NVMe over Ethernet networks will help these storage networks dominate storage networks in the coming years.
In fact, Western data announced a hard drive.
Storage box based on NVMe
Ethernet connection as part of its OpenFlex architecture.
Big Data applications are driving a new computer architecture in which processing is closer to memory rather than moving a lot of data from storage and memory to the processor.
This leads to a new combination of storage with a special purpose processor (commonly referred to as an accelerator.
For example, Samsung has launched a smart nvme ssd, which includes an ARM processing core focused on database applications.
This SSD is an example of computational storage with processing included in the storage device.
Several other companies are introducing computing storage devices.
SNIA established a computing storage technology working group in the fall of 2018 as part of the Solid State Storage Initiative (SSSI).
Members include ARM, ideticom, Inspur, Lenovo, Micron Technology, NetApp, NGD Systems, Nyriad, Samsung Electronics, ScaleFlux, SK hynix, Western DigitalRISC-
V is a free and open computer standard dating back to 1980s. The RISC-
Founded in 2015, the V Foundation has more than 100 member organizations working together on software and hardware innovation, which the foundation believes will pave the way for computing design and innovation over the next 50 years.
Supporters of streamlined software
V says it makes the diversity and workload of big data and fast data applications soar in core data centers as well as remote and mobile systems at the edges.
It provides an alternative to the current, standard, universal computing architecture.
Companies like Western data are planning to use
The V core in the storage device provides a very flexible open application architecture for computing storage.
The way to move some flash management to the SSD of the host is called Open-channel SSD. In an open-
There is no firmware flash translation layer for the channel SSD (FTL).
But the management of entities --
The status is stored in the computer's operating system. The Linux 4.
4 kernel is open-
Channel SSD built in accordance with NVMe specifications.
Have a community that supports open developmentchannel SSDs (
This approach is attractive for some very large scale SSD applications.
Another way to handle closer memory is called in-
Memory processing.
At 2018 ieee iedm Conference, companies working in this field published several papers.
Various emerging non-
Consider using the memory that contains the data to be processed to include the processing core for special purposes.
This idea can be extended to the actual use of non-
Volatile as a memory and processing device.
This method, usually using a neural morphological calculation method, is processed using a device similar to the nature of brain neurons, and has been using magnetic random access RAM (MRAM)
Resistance RAM (RRAM)
Phase Change Memory (PCM)technologies.
At the IEEE restart computing conference, various methods of packaging logical memory using special packages of multiple connected chips are envisaged, as shown below.
These heterogeneous integration methods may become an important element in embedded and enterprise architecture in the future.
There's a new
Unstable memory technology is coming out of the lab into the real world.
These technologies include the memory mentioned above, as well as the iron body memory and other more exotic memory technologies.
Intel's Optane Memory uses 3D XPoint (
Phase Change Memory)
It is developed with Micron Technology.
Intel announced in 2018 that it will launch the DIMM-
In addition to the current NVMe Optane SSD, it is also based on the Optane Memory module.
As a memory solution associated with the next generation server processor, these products should be put into use in 2019.
All semiconductor factories have said that they will launch embedded non-
As of 2019, volatile memory, including MRAM and RRAM.
In fact, a company called Gyrfalcon recently announced that TSMC used 40 mb of 22nm embedded MRAM for an AI chip they made in 2018. These new non-
The memory of instability will be replaced, perhaps even higher --
Level buffer memory (L4 and L3)in processors. Use of non-
Volatile memory will help reduce energy consumption and speed up startup for products that use these embedded memory.
According to Khao Fulin, No.
By 2023, the volatile memory market could exceed $ 6B.
The following figure shows the high, medium and low forecasts for total shipments of emerging memory annual shipments as of 2028.
Although SSDs are heavily used as primary storage in the data center and gradually replace hard drives in many client applications, the total capacity stored on hard drive drives and tapes will continue to grow, to meet the needs of fast storage and bulk storage.
The following figure gives our forecast of annual capacity shipments.
The next generation of applications will drive demand for all types of storage devices.
To meet the needs of fast storage and bulk storage, the next generation of applications will drive demand for all types of storage devices. Solid-
State memory and storage enable people to process and use explosive data generated by people and machines in new ways.