What are the Types of Computer Memory?

What are the Types of Computer Memory?

Memory is one of the foundations of human cognition. A man from ancient times tried storing information. At first it was drawings and diagrams, then the speech appeared, which helped to convey information, and later, writing was produced – the record of speech. Finally, in the 20th century, computing systems with their own binary "language" appeared, which could process and store information in huge quantities. Computing systems use a variety of memory, or storage devices. Some are needed for long-term storage others need to be overwritten at high speed to pass the information to processing. Let's take a look at the current types of memory for computing devices and what to expect in this area in the future.

Current types of data media

Need to say that different types of semiconductor memory are mainly used now, but it is worth starting with data storages popular in the past, but less and less used now.

A few decades ago, the main method of long-term information storage was magnetic recording. Data is recorded using heads that generate a magnetic field and change the magnetization in a magnetizable material, which is usually applied to the disk, or tape. The latter was previously widely used to record audio and video information, but is now used in magnetic stripe cards as access keys or bank cards. But even in these applications, magnetic tape is replaced by more convenient RFID technology. Note that sometimes there are reports of attempts to make media on magnetic tapes with large amounts of data, but such devices will have a very long time to access information, so they are only useful for archives and backups.

Magnetic disks, or rather hard disk drives (HDD), are still widely used in computer systems, servers, and network attached storages (NAS), although they are gradually replaced by silicon-based flash memory. The advantage of HDD is the lower price of 1 GB of storage and greater durability with frequent overwriting. The main disadvantage of HDD is the presence of precise mechanics for positioning the magnetic heads and spinning the disks themselves, which leads to problems with mechanical stability, noise at work, physical size and weight. These settings are especially important for mobile devices, so hard drives have completely disappeared from laptops and are not used in other mobile electronics. 

The next important type of media is optical disks. The data on the disks are recorded as dark spots (which reflect light worse) on the surface of the reflective material. Reproduction of information is by measuring the reflectivity from the surface using a laser and a photodiode. The information layer is closed with a transparent protective material. Information is recorded on optical disks with the help of a higher-power laser, which burns the reflective layer, which darkens. There are optical disks with the possibility of multiple recordings (up to tens of times) with "discoloration" of the surface by the same laser. The main problem with optical disks is slow recording. They are almost impossible to use for the temporary storage of information. But the tremendous advantage that once made optical disks a mass phenomenon is the ability to "stamp" the same disks with the same information in large numbers and extremely long disk life. Today, almost all applications of optical disks are video, audio and game distribution. Even the use of digital sales of such products has not completely supplanted optical disks, as many people buy disks for collections, and with a view to future resale. However, from year to year the percentage of digital sales is growing because it is more profitable for manufacturers. 

Modern optical disks have two layers on which data is placed, and writing and reading is a laser with different wavelengths. Such disks hold up to 120 GB of information, which is close to the limit for this type of recording, because the size of the "spot" must be significantly larger than the wavelength. Holographic recording can become a development of optical recording technology. Here, the recording will be made in the volume of the transparent layer at points where two coherent laser beams from one source will be focused at once. Such disks will be read using a CCD camera. The recording density of information in the case of holographic disks can reach 30 GB per cubic millimeter, or 10-30 TB per disk, and the lifetime can reach 50-100 years. However, due to the complexity of such technology there are no commercial applications yet, however, it is very promising for backups.

Of course, semiconductor memory is the most interesting in terms of Internet of Things and computer technology in general. There are many types of silicon memory, depending on the structure of the cells, the principles of memorization and production. The fundamental difference between current semiconductor memory and disk media, which records in sectors on various tracks and requires spinning the disk to begin reading or writing, is that all cells have the same access time. This property has led to the fact that almost all semiconductor storage devices are associated with the abbreviation RAM (Random Access Memory). Another advantage of semiconductor memory is the petite size of the media, which decreases from year to year according to Moore's Law, as well as the ability to produce memory modules on the same substrates as computing modules, which leads to even greater integration and reduction of total power consumption. Like all types of memory, semiconductor one can be divided into two areas: energy-dependent memory, which can't work without added power, but has a high switching frequency and bandwidth (cache, main memory), and non-volatile memory, where information remains stable in time.

Energy-dependent memory

The basis of most modern microelectronics is the metal-oxide-semiconductor field-effect transistor (MOSFET), or simply MOS transistor; of course, storage devices are no exception. Depending on the scheme and the principle of operation energy-dependent or volatile memory is divided into dynamic and static.

Static random-access memory (SRAM). One SRAM cell (one bit of memory) requires 4, 6, 8, or 10 transistors, now, of course, using a MOSFET. A simple cell circuit is a binary trigger consisting of two cross-coupled inverters (1, 2, or 3 transistors each) and two additional access transistors. Such triggers do not need to be updated, but cannot store information without constant power. The great advantage of SRAM is fast access to memory cells, great switching speed and low power consumption. However, such memory is very expensive, because the cells consist of numerous elements. Thus, the main uses of SRAM are in-processor caches, HDD and router buffers, buffers in monitors and printers. SRAM is also used in low-power microcontrollers and field-programmable gate arrays (FPGAs) typically up to 128 KB.
Static random-access memory (SRAM) architectureDynamic random-access memory (DRAM). A DRAM cell consists of a capacitor and a single transistor through which the capacitor is charged (state "1") and discharged (state "0"), and cells are combined into large arrays. Such a simple scheme allows getting a huge density of cells per area unit, but it leads to other problems. Due to the simplicity of the circuit, the capacitor is gradually discharged, so it must be periodically recharged, which leads to unnecessary energy consumption. In the integrated circuit chips of memory, DRAM cells are grouped into rows, rows are grouped into pages, and pages into banks. In most DRAMs, only rows (not cells) can be read. This structure of modules requires the availability of a memory controller: a digital circuit that manages the flow of data going to and from the chip, it also implements virtual addressing to make available an operating system data to process. 
Dynamic random-access memory (DRAM) architectureBecause of additional controllers, delays occur between the command to read or write data in DRAM and its execution, which is measured in clock cycle. There are four types of delays, which are marked on the body of finished microchips: column address strobe latency (CL), row address to column address delay, row precharge time and row active time. Due to its characteristics, DRAM is used as computer main memory (as well as for servers, single-board computers, controllers, where it is called embedded DRAM or eDRAM, smartphones, etc.). It is worth noting that the market for DRAM chips alone is now more than 20 billion. 

Comparison SRAM vs DRAMA new type of memory – Thyristor RAM, or T-RAM, may be promising. It uses the area of negative resistance of the thyristor to ensure the storage of the bit information. This reduces the number of electrical elements required per cell compared to SRAM without the need to recharge, as in DRAM, but there are still problems with the commercial production of such memory.

Thyristor RAM (T-RAM) architecture Non-volatile semiconductor memory

The most common non-volatile memory is now floating-gate MOSFET based erasable programmable read-only memory (EPROM), better known as flash memory. The modern flash memory market has already reached 80 billion and continues growing. Flash-memory based storage is used in mobile electronics (laptops, tablets, smartphones) as a permanent repository, in computers and servers as a solid-state drive (SSD), in memory cards and, of course, in controllers and microcomputers.

Flash memory works by changing the level of charge stored internally behind the gate of the MOSFET. The gate is created with a special "stack" designed to hold the charge in "traps" based on a floating gate. The presence of a charge inside the gate changes the threshold voltage of the transistor, making it higher or lower, corresponding to "1" or "0". This type of memory cell is called a single-level cell (SLC), also can be a multi-level cell (MLC) with 2 bits per cell, or four voltage levels, TLC (3 bits per cell), QLC (4 bits per cell) and so on. Changing the state of bit's requires the accumulated charge to be dumped, which in turn requires a relatively high voltage to "pull" electrons from the floating gate. This flash is provided by pumping the charge, which takes some time to accumulate power, typically 100-300 μs for a block of data, which is about 10,000 times of the typical write time 10 ns for SRAM, for example.

Flash memory architecture

This method of erasing data is not only the reason for the memory type name, but the most important problem. First, flash can erase only a block of memory, so, if you need to change a part of the data from the block, you need to read it, erase it by flash and write it again. Second, erasing leads to the step-by-step degradation of memory cells, due to even the minimal structure difference of cells on erasing, a different charge will remain, and after many cycles, their information will not be correct. Especially this concern to multi-bits cells, so they can often survive only 3-5 thousand rewrites. However, the incredible popularity of flash memory, despite all the difficulties associated with the density of recording compared to other media. Modern flash memory is not flat, but a multilayer structure with a number of layers of three hundred, each of which contains cells of transistors with floating-gate, and therefore the recording density is hundreds of times greater than the flat structure.

The main competitor of flash memory is Phase-change memory (PCM or PRAM) a type of non-volatile random-access memory based on unique properties of chalcogenide glass, which can be in crystalline (conductive) and amorphous (dielectric) states. The transition between states is due to changes in temperature. Immediately clear, when such media is heated (for example, when soldering chips), the record will be erased. Also, a problem is the large minimal size of chalcogenide amorphous structures (relative to the size of modern transistors). However, the advantage of PRAM is the high speed of writing and reading compared to DRAM, and durability: the cells can relive tens of millions of erasing cycles. The most famous and commercially successful PCM is the proprietary 3D XPoint technology. Due to the complexity and cost of technology, as well as the impossibility of creating multilayer structures, such memory is very expensive and has a low recording density compared to flash memory. However, the use of such media is advisable instead of DRAM, which will help, for example, reduce power consumption by mobile devices, or allow turning on laptops or PCs almost instantly.

Phase-change memory architecture There are many other types of promising media, such as Resistive RAM (RRAM) that works by changing the resistance across a dielectric solid-state material. Under the action of charge, conductive filaments appear or disappear in the dielectric. This type of memory has not reached the stage of commercialization yet, but is interesting from another point of view. The RRAM cell is actually a memristor - the fourth type in a theoretical quartet of fundamental electrical components. Its use can radically change our understanding of electronics and data processing principles, as it will allow allocating memory in computing cores without the use of cache. This approach is promising for high-performance neurocomputers.

Resistive RAM (RRAM) architecture Conclusions

Data media and storage devices have come a long way and now use various physical properties and materials. In some cases, miniaturization and attempts to write increasing amounts of information are already facing physical constraints, but continue to grow, and the growing popularity of the Internet of Things and edge computing poses new challenges for data processing and storage device manufacturers.