Home  Memory

Jul 25 2019

Future Trends in NV and NVM Memories

Introduction

Non-volatile memory (NVM) is a computer memory that does not disappear when the current is turned off. In non-volatile memory, depending on whether the data in the memory can be rewritten as a standard at any time when using the computer, it can be divided into two major categories, namely ROM and Flash memory.

Read-Only Memory (ROM) is a type of semiconductor memory that is characterized in that once data is stored, it can no longer be changed or deleted, and the content does not disappear due to power off. In electronic or computer systems, it is usually used to store programs or data that do not need to be changed frequently, such as early home computers such as the Apple II supervisor, BASIC language interpreter, and hardware dot matrix fonts, personal computer IBMPC/XT/AT The BIOS (Basic Input Output System) and the IBM PC/XT BASIC interpreter, as well as the firmware in various other microcomputer systems, are stored in ROM.

Flash memory is a form of electronically erasable programmatic read-only memory that allows memory to be erased or written multiple times during operation. This technology is mainly used for general data storage and exchange of data between computers and other digital products, such as memory cards and USB flash drives.

In this video from the 2015 OFS Developer's Workshop, Dave Akerson from Intel presents: NVMe Introduction and Tutorial.



Article Core

 NV and NVM memories

Purpose

Introduce the future trends in nv and nvm memories

Application

Semiconductor Industry.


Catalog

I

Introduction

II

Overview

III

RAID Storage

IV

RAM is Important

V

NV Solid State Memory

VI

The Future of Flash Memory

VII

Mass Storage

VIII

Keep Interconnected


Overview

In general, the semiconductor memory that we can retain after the data is powered off is called “non-volatile (or non-functional) random access memory”—Non-Volatile Random Access Memory (NVRAM), like DRAM and SRAM. The memory is called VRAM. In fact, strictly speaking, non-volatile memory should be called NVM (Non-Volatile Memory), because some memories have very weak random access capabilities, but in this article, for convenience, they are collectively called NVRAM, which is also due to Both future and future non-volatile memories have good or even very good random addressing capabilities, and the focus of this article will be around them.

Why do we talk about NVRAM today - this less familiar thing? Because NVRAM is already one of the most important memories in the present and future, its limelight has even overshadowed the traditional DRAM and SRAM we are familiar with.

Due to the VRAM characteristics of DRAM and SRAM, there are certain restrictions in the field of data storage. In fact, each type of memory has its own limitations, but the VRAM feature makes these memoryes that are usually responsible for important data transfer tasks in some cases. In the hard disk field, there is a write cache technology. The application process is when the host writes data to the hard disk. When writing to the write cache, it means that the write is completed for the host side, so that the latter work can be performed, but If the power is suddenly turned off at this time, it means that the data in the write cache is lost and not actually written to the disk. Therefore, in the field of network storage, many disk arrays themselves have an uninterruptible power supply to ensure the security of data in the array cache (generally DRAM of GB capacity).

On the other hand, digital handheld devices quickly became popular after the 1990s. These small devices are like a PC, so they are all well-organized. At this point, you must have a memory that can take the role of the hard disk in the PC for the handheld device. However, the conventional magnetic disk (Magnetic Media) requires a large, complex and fine mechanism to drive the magnetic head and the disk, which is liable to be abnormal due to mechanical wear or external impact. In addition, disk drives need to consume considerable power to maintain the operation of the motor and electronic drive circuits. Even though there are many power-saving features (especially laptops) in today's computer designs, mechanical storage devices still account for a large portion of the computer's power consumption. Part. Obviously, this is not very suitable for a small handheld device. Although there are now 1 inch and 0.85 inch hard drives, the use of handheld devices is still immature. It is currently only applied to devices of the high-end digital camera (DC) level. However, the development of DC mobile phones, PDAs, and portable DCs increasingly requires large-capacity memories.

In contrast, DRAM and SRAM, although small in size, are not affected by collisions, are very fast, but do not have NV characteristics, so they can only act as traditional memory in handheld devices. Obviously, there seems to be a vacancy here, and this vacancy is exactly what NVRAM can make up. It has the physical advantages of DRAM and SRAM, and also has the permanent (relatively) storage characteristics of the hard disk. Therefore, under the push and demand of this industry, it has developed rapidly, when the transmission is getting closer to DRAM and the single chip capacity. When it was significantly higher than DRAM, people realized its great potential. If you still don't know, don't say anything (hard disk, optical drive, graphics card, motherboard, etc.), you can see how many PDAs are in use around you, how many people are using mainstream DC, how many Using a mobile phone? There must be NVRAM in their "body".


RAID Storage

In the future, research on DRAM and NVRAM memory seems to have a convergence trend. The reason is that NVRAM's appetite is getting bigger and bigger. When the speed and cost are similar, the unique ability of NVRAM will be powerful for DRAM and SRAM. Challenger. Therefore, we are paying attention to NVRAM today, and we are also paying attention to an important branch of semiconductor memory, and it will become more and more important.

Every processor requires memory. Some systems may use only one type of storage, but more commonly, a layered storage system, such as a server's RAID storage system. Each type of memory makes a different contribution to the system, such as large capacity, fast access, or non-volatile.

RAID Storage System 

The memory combination is changing with the advent of new technologies and improvements of existing technologies. What has not been heard before has become a standard. For example, netbooks are now only available with solid-state memory, for that has a longer battery life due to lower power consumption of solid-state memory. 

The demand for larger capacity memory requires trade-offs and different implementation methods. For example, multi-level cell (MLC) flash can provide more capacity than single-level cell (SLC) flash, but at the expense of performance and hardware lifetime.

Also, the hard drive size is decreasing. The 2.5-inch hard drive is a huge improvement over the old-fashioned, full-height 5.25-inch hard drive, and the 2.5-inch drive has more capacity and faster response times. RAID has also gained wider use due to it has provided more reliable storage.


RAM is Important

RAM is the center of commercial computing devices. Currently, non-volatile core memories are often substituted by volatile memory SRAMs and DRAMs. However, new non-volatile memory technologies such as ferroelectric memory (FeRAM) and magnetic RAM (MRAM) are expected to change this situation.

Standalone SRAM chips are still in use, but most SRAMs are typically integrated into the microcontroller chip, providing different functions from file registers to multi-level cache. The main advantage of SRAM is its high performance, but the large chip area and the high power consumption are its desperate disadvantages.

DRAM makes things more interesting. On-chip DRAM is becoming more and more common, although different semiconductor technologies that implement DRAM and logic tend to make DRAM and logic into two separate chips. In addition, the higher capacity of DRAM has led designers to take DRAM out of the processor chip. The result is that the designer can choose how much capacity to provide, or the end user can even add their own memory.

Embedded designers face many other challenges when choosing DRAM because the microprocessors like DRAM used vary widely in performance characteristics. Embedded designers also need to consider the product life cycle, because PC users tend to pursue the latest, best, and lowest cost memory when choosing a memory. The trend toward virtualization is driving the need for higher-density storage, and the motto is that the memory is never enough.

SDRAM is widely used in low-end applications, at least in embedded applications. SDRAM has always been plentiful and inexpensive, and one of its main advantages is its simple interface requirements. The slower speed of SDRAM is an advantage for designers compared to the DDR2 and DDR3 memories used in today's large PC systems, especially when trying to match slower processors. Compared with DDR2 and DDR3, the two major drawbacks of SDRAM are capacity and efficiency.

Another issue that microprocessor designers are experiencing is speed. Raising the upper speed limit usually means raising the lower limit at the same time. This is not a problem when paired with the latest x86 GHz multi-core processors developed by AMD, Intel and VIA (VIA), but when trying to support a 200MHz processor, the problem comes.

Of course, the processor clock can be increased, but this will increase the cost and power consumption accordingly. These two indicators are absolutely what we try to reduce. Almost all microcontrollers (MCUs) can be paired with SDRAM. Some MCUs can be paired with DDR2, but few MCUs can handle the high speed of DDR3.

DDR2 is currently the most demanding. It is widely used in servers, PCs and laptops, but these products are rapidly turning to DDR3. In the coming period, although the supply of DDR2 began to decline and prices began to climb, DDR2 will still be favored by embedded systems. This won't happen overnight, but it's a development trend. The challenge in the embedded market is how to make low-end MCUs meet the performance requirements of DDR2.

Samsung's new 16GB DDR3 is aimed at server motherboards, which are usually designed to support only DDR3 memory. When using these new modules, the server board can support 192GB of DDR3 with a transfer rate of 1333 Mbps and a 60% reduction in power consumption over DDR2. Many high-end motherboards have chipsets that support both DDR2 or DDR3. Chipsets that only support DDR3 are typically smaller and more efficient.

DDR3 Memory 

The Z-RAM single-transistor memory technology developed by Innovative Silicon has better scalability and smaller chip area than existing DRAM technology. Both Hynix and AMD have licensed Z-RAM technology, but their application goals are different. Hynix may integrate it into its main memory, and AMD wants to use it as a large on-chip L3 cache.

Serial port memory is designed to introduce high-speed serial interfaces into memory. In theory, it will reduce the number of pins required for memory by 40% and provide throughput of 3.2 to 12.6GBps. Its initial application target was multimedia mobile devices, which had very tight PCB space and required low power consumption.


NV Solid State Memory

DRAM is volatile, but non-volatile (NV) memory is always part of the system solution. In recent years, non-volatile solid-state memories have undergone tremendous changes, capacity is increasing, and costs are falling. At present, many non-volatile memories have been widely used, from flash memory to MRAM to FRAM.

ROM is a well-known for non-volatile memory technology that is favored in standard microcontrollers. Because ROM is the most efficient non-volatile storage technology, it has always had a place in custom chips. Unfortunately, the content in the ROM cannot be changed as in other non-volatile storage technologies discussed herein.

An example of a ROM application is Luminary Micro's LM3S9000 microcontroller, which has a runtime library that provides StellarisWare library services. It is different from a typical ROM-based custom microcontroller that includes the entire application. In the MCU of Luminary Micro, the main application using the ROM code is stored in another non-volatile memory. The ROM may only contain boot code, which allows the main application code to come from different sources, including from the network.

Flash memory can cover a wide range of solutions. FRAM and MRAM have similar characteristics, and their prospects are very bright, and are currently used mainly in important gap applications.

These non-volatile memories can effectively replace SRAM because they have the same operating speed as SRAM, and they do not have the write limit problem that flash memory faces. This allows them to be used as primary and secondary memories. Their capacity is increasing and costs are falling, although they still lag behind SRAM and flash. This led to some interesting combinations, such as the RAID controller mentioned earlier.

Why we study the non-volatile memories?For example,the CPU is getting faster and faster, but the external storage (disk or SSD) speed is much slower than the CPU speed. Many services that require high data consistency, such as traditional databases, need to write data to external storage through the log for each operation. Although the CPU and DRAM are faster, the overall performance is still low due to the slow external storage speed. Imagine, if our memory capacity becomes as large as the hard disk, and the data will not be lost when the power is turned off, then the short board of the whole system will be filled up. It is fast to run the business on such a system.

FeRAM supplier Ramtron's 8051-based VRS51L3x.x.x microcontroller family integrates 64kB flash, 4kB SRAM and FeRAM up to 8kB. Flash memory is used to store program code and long-term, less frequently changed data. SRAM and FeRAM are used to store read/write data, where FeRAM is used to store data that requires non-volatility.

 16-Mbit nvSRAM 

FeRAM and MRAM also offer pin-compatible models that replace SRAM and flash. Everspin's MR2Axx MRAM product line is pin compatible with standard 8- and 16-bit SRAM devices. These devices also offer BGA packages with 35ns read/write speed and extended industrial temperature range. Everspin's up to 512kB MRAM devices have been used in Emerson Network Power's Freescale MPC864xD-based MVME7100 single-board computer .

Everspin 512kb MRAM 


The Future of Flash Memory

Flash is the current mature technology. Flash memory in most stand-alone flash products has a higher density than flash memory integrated in the MCU because the flash integrated in the MCU must use the same process as the logic implementation.

Stand-alone flash memory also comes in different formats, from chips to removable devices such as CF cards, SD/XD, MiniSD, MicroSD, Memory Stick and USB. These products have been used in embedded applications such as WinSystems' 16 GB industrial grade CF card. Its dual channel operation supports a sustained read transfer rate of 40MBps and a write rate of 30MBps.

WinSystems' 16 GB industrial grade CF card 

There are more options for embedded applications. Modules that can be inserted into the IDE header are a common alternative to hard drives. Initially, the capacity of these flash memories was small. However, their capacity has now increased dramatically, making these flashes not only useful for boot code storage in many applications, but also a complete replacement for hard drives.

Western Digital Solid State Storage (formerly Silicon Systems) is a supplier of flash drives using the Small Form Factor (SFF) SIG Silicon Blade form factor. Its Silicon Drive Blade is rugged and rugged, and is a replacement for the 10-pin module from Western Digital Solid State Storage. It has multiple supply channels that plug into the 10-pin header on most PC motherboards.

Western Digital Solid State Storage 

Flash memory shapes are often less important when making decisions than other technology options. Is it NAND, or is it NOR? Is it SLC, or is it MLC? These technical choices make designers have to face a series of trade-offs. No single product meets all application needs. In fact, in some of the more demanding applications, multiple flash technologies are requested.

We can understand some of the trade-offs that designers must make from some of the Toshiba products' common metrics. For example, the erase speed of NAND is 2ms, and the NOR is 900ms. On the other hand, the capacity of NOR is four times that of NAND, which can reach 256 Mb, and it is still increasing. When using a 103MBps clock, the read rate of NOR is at least 4 times that of NAND. But the write rate of NOR is at 0.5MBps, while that of SLC NAND is 8MBps.

SLC and MLC have similar trade-offs. MLC has a higher density but writes much less often. All flash technologies have limitations, which makes alternative technologies such as MRAM and FRAM very popular. If MRAM and FRAM technology can provide near or over the capacity of flash memory at a similar price, the layout of the memory market will change significantly. Unfortunately, this is unlikely to occur in the short term.

This means that wear leveling techniques are becoming more and more important, especially considering the limitations of MLC technology in this regard, as well as its significantly larger capacity. The target period for hard disk replacement is 5 years. While this is a long enough enterprise-level solution, it may not be appropriate for embedded applications with longer lifecycles. This means that designers must pay more attention to specifications than ever before. 

Designed specifically for MLC flash, SandForce's SF-1500 SSD controller provides a random read/write rate of at least 5 years, 30k IOPS (input/output per second), and 250MBps continuous read. / write operation rate. For a hard disk, it is equivalent to a ratio of 5k IOPS/W to 20 IOPS/W.

SandForce's SF-1500 SSD controller 

The DuraClass technology used by SandForce also uses a redundant array of independent silicon components (RAISE), which is essentially a RAID with chips. Combined with advanced dynamic wear leveling and advanced error correction coding (ECC) support, it allows SandForce-supported Solid State Drives (SSDs) to meet the performance and longevity requirements of enterprise storage.

Alternative technologies will have difficulty meeting these requirements unless a similar approach is taken to circumvent the limitations of MLC flash. For example, to guarantee a five-year lifespan, many alternative technologies mandate the maximum number of writes per day. SandForce supports single-chip controller solutions in 512GB, 1.8-inch SSDs.


Mass Storage

SSD is one of the mass storage solutions. SSDs have wiped out the 1-inch hard drive market and are gradually increasing their share of the 1.8-inch, 2.5-inch or even 3.5-inch hard drive market. In terms of form factor, SSDs also vary greatly, and they do not follow conventional hard disk configurations. This is also why SSDs can be easily placed on a circuit board, but hard drives are difficult to do.

Despite this, the SSD is still not as good as the hard drive in terms of the maximum capacity that can be achieved. From the price/gigabyte perspective, the hard drive is also a winner. The SSD replaces the hard drive's transition point and keeps moving, but this simply means that designers and users have more options.

1.8-inch hard drives are most popular with mobile devices. It is also here that consumers have become more difficult to choose between flash and hard drives. While this is easy for designers, both SSDs and hard drives offer this form factor. (But the trade-off between price and capacity still exists.)

Most of the action takes place in the 2.5 inch field. It includes an external hard drive like Fujitsu's 500GB Handy Drive.

Fujitsu's 500GB Handy Drive 

Since a large number of hard drives can be easily installed in a 1U rack, the form factor has also significantly affected the design of the server. More importantly, this number greatly exceeds the minimum requirements for RAID configurations, which has contributed to the growth of the controller market. An eight-drive RAID system is no longer new. Instead, it has become a standard choice, and even more drives have appeared in high-end storage systems.

Compared to the capacity of a 3.5-inch hard drive, 2.5 inches is still a small one. But for RAID systems, the size of the header is not everything, because for smaller hard disk configurations, the number of system rebuilds is less.

Don't ignore the 3.5-inch market. Like Seagate's Barracuda LP hard drive, which has 2TB of capacity, it focuses on meeting the capacity requirements of digital video recorders for video storage. If the film company recognizes that this growing storage capacity will give them opportunities, the 3.5-inch hard drive market will be very hot. If so, the 3.5-inch hard drive is likely to be in short supply.

Seagate's Barracuda LP hard drive 

RAID will continue to work with 3.5 hard drives, especially in consumer electronics applications. However, it is important that users do not understand this. It can be easily explained why it is necessary to provide consumers with more storage capacity, and even make it possible to reduce the capacity with RAID to improve reliability. But to understand the difference between RAID 1 and RAID 5 is another completely different thing.


Keep Interconnected

Any description of the memory is incomplete without mentioning the growing importance of the interconnect. For consumer-centric products and a variety of embedded applications, this means USB and SATA. USB is an indirect interface to the hard disk and may become a direct interface to the flash drive.

External SATA or eSATA quickly appeared in many products, including external drives, but it will complement, not replace, USB. USB 3.0 will appear in time to meet the needs of higher throughput drives. Although currently, Hi-Speed USB 2.0 is sufficient for its 480 Mbps transfer rate.

SAS and Fibre Channel (Fibre Channel) will appear in the enterprise storage market. Fibre Channel systems often include SATA or SAS drives, and may or may eventually include SSDs.

As far as storage is concerned, there are more options than ever before, but making choices is not easy because there are multiple alternatives


You May Also Like:

What is Computer Memory? Types of Computer Memory

What Are the New Memory Technologies and What Problems Can They Solve?

The Function, Structure and Working Principle of Cache Memory


0 comment

Leave a Reply

Your email address will not be published.

 
 
   
Rating: