Fullscreen HTML5Med Res (??MB)HTML4

Data Storage Systems Part 2: by Dr David Maddison Last month, we covered older storage systems like core memory, magnetic tape, floppy disks and optical discs. This follow-up article will describe modern storage technologies like hard disks, flash memory and SSDs, as well as possible future storage systems like 5D optical, holographic and DNA storage. W hile SSDs have displaced hard disks in many applications, especially for portable computers, mechanical hard disks are still widely used. That’s due to their lower cost and higher storage density, although flash may catch up eventually. Advances in mechanical hard disk storage are still being made, though. We will now look at how both technologies have evolved over time and where they are now. Hard disks/drives Hard disks (or hard drives) store data on internal rotating discs (‘platters’) coated with a thin film of magnetisable material. Movable heads magnetically read and write data on the individual platters (usually on both sides at once). Individual data bits are represented by the magnetisation of tiny magnetic domains (see Fig.31). Modern disk heads ‘fly’ on a thin layer of trapped air just above the platter surface. If the heads ever contact the surface, due to a physical shock or other reasons, it is known as a “head crash”; data loss and head damage can occur. Modern drives try to avoid head crashes by parking the heads in a special zone when the power is off, no data is being accessed or if they detect sudden acceleration. The IBM RAMAC (Random Access Method of Accounting and Control), introduced in 1957, was the first commercial computer with a hard disk drive of about 3.75MB. According to the RAMAC operations manual (siliconchip.au/link/abrw), THE IBM RAMAC is built around a random-access memory device that permits the storage of five million characters of business facts in the machine. In effect, the machine stores the equivalent of 62,500 80-column IBM cards. The Model 350 drive (Fig.32) had 52 platters, of which 50 contained data on 100 surfaces, and a read/write head unit on a moving arm that held two heads. You can see a video of it working at https://youtu.be/aTkL4FQL2FI The Bryant Chucking Grinder Company started developing a disk drive unit in 1959, resulting in the introduction of the 4000-series in 1961 (see Fig.33). It contained 26 horizontally-­ mounted discs 99cm in diameter spinning at 1200 RPM. The 205MB capacity was enormous for the time. You can see their 1965 product brochure at siliconchip.au/link/abrx IBM introduced the Model 1311 disk drive in 1962, which was about the size of a washing machine. It had a removable ‘Disk Pack’ containing five 35.5cm platters with ten recording surfaces that spun at 1500 RPM. The Pack weighed 4.5kg. It stored 2 million characters, equivalent to approximately 25,000 punched cards. In 1973, IBM introduced the “Winchester” disk drive, with 360mm platters, which did not have a removable Table 1: hard drive evolution since 1957 1957 1970 1980 1990 1995 2000 2005 2010 2015 2020 Capacity 3.75MB 29MB 5MB 120MB 4GB 80GB 500GB 3TB 10TB 20TB Volume 900L 768L 2.4L 2.4L 0.39L 0.39L 0.39L 0.39L 0.39L 0.39L Weight 900kg 360kg 2.3kg 2.9kg 1.5kg 0.7kg 0.7kg 0.7kg 0.7kg 0.7kg Access time 600ms 50ms 85ms 28ms 8.5ms 8.5ms 8.5ms 8.5ms 8.5ms 8.5ms $6,000,000 $1,500,000 $7,875.00 $250.00 $2.80 80¢ 8¢ 6¢ 2.5¢ 20kb/cm2 125kb/cm2 2Mb/cm2 50Mb/cm2 2Gb/cm2 13Gb/cm2 97Gb/cm2 128Gb/cm2 180Gb/cm2 7000 11,000 40,000 250,000 500,000 1,000,000 2,000,000 2,500,000 2,500,000 US$/GB $9,200,000 Areal density 309b/cm2 MTBF (hours) 2000 16 Silicon Chip Australia's electronics magazine siliconchip.com.au Fig.31: magnetic domains representing data bits on the platter of a 200MB hard disk. Source: https://w. wiki/8XxE (CC BY-SA 3.0). Fig.32: the 3.75MB IBM Model 350 disk drive used in the 1956 IBM RAMAC 305 computer. Source: https://w.wiki/8XxG (CC BY-SA 2.5). Fig.33: the Bryant Chucking Grinder Company Model 2 disk drive, or “disc file” as it was called. Source: www. computerhistory.org/timeline/1959/ Disk Pack. Then, in 1978, IBM introduced the “Piccolo” Model 0680 with smaller 20cm (8in) platters to replace 8in floppies. Over time, hard disks shrank along with floppies, first to 5¼ inches (133mm), then 3.5in (89mm). That final size is still widely used today. At the beginning of the 1980s, hard disks were uncommon for PCs and very expensive, but they reduced in price dramatically toward the end of that decade. Improvements in capacity, density, speed, size, price, reliability and other factors are shown in Table 1. The first hard drives (for mini and mainframe computers) with a standard interface were the Sperry Univac RP01, RP02 and RP03 drives (sold under several names). The RP02 was released in 1969 with a 20MB capacity. The interface design was not made proprietary, resulting in it becoming widely used. Early hard disk interfaces on PCs had a controller card and two cables, one for control and one for data. Popular early defacto standard interfaces were the ST506 and ST412 from Seagate (named after specific hard disk models that used them). ST412 was a refined version of ST506 and was used on the IBM XT. Both used MFM (modified frequency modulation) encoding, but an extended version of ST412, ST412HP, used RLL to give a 50% increase in capacity. I once had a 40MB Miniscribe 3650 hard disk with an MFM controller card, and I swapped the controller for an RLL (Run Length Limited) card, reformatted the disk and achieved a 60MB capacity. Following these interfaces came IDE (Integrated Drive Electronics), also known as Parallel ATA (PATA), which was developed by Western Digital and Compaq and introduced in 1986. It became the ATA-1 standard that virtually all PCs used in the late 1980s and early 1990s. Communication was over a 40-wire ribbon cable with IDC connectors at each end, while power was supplied separately. Enhanced IDE or EIDE was introduced in 1994, closely related to the ATA-2 standard. Further developments of ATA were ATAPI (for devices other than hard drives), ATA-4 with UDMA (Ultra Direct Memory Access), then Ultra ATA variations up to ATAPI8. Later versions of ATA used 80-wire shielded ribbon cables but with the same 40-way IDC connectors. SCSI was a general-purpose interface designed for various devices, including hard disks. It existed concurrently with ATA; it was more flexible, reliable and faster but more expensive to implement, so it was used in higher-end computers such as servers. Current hard drive interfaces include: • Serial ATA (SATA), released in 2003 to replace the IDE/PATA interface, using much thinner cables with fewer conductors. • SAS (Serial Attached SCSI), introduced in 2004, mainly for enterprise computing. It uses cables and connectors similar to SATA. • The M.2 interface is designed for solid-state drives (SSDs). It can utilise a SATA link or the faster PCIe bus (Peripheral Component Interconnect Express) with the NVM Express (NVMe or nonvolatile memory express) communications protocol. • mSATA (mini-SATA) is designed for space-constrained applications for SSDs, but today, M.2 is more likely to be used for such applications. • U.2 (SFF-8639) is designed for enterprise applications where very high performance is required. It uses the PCIe bus and can utilise the NVMe communications protocol. • FC (Fibre Channel) was introduced in the 1990s but has been adapted to SSDs today and is used in enterprise applications. Since 2010, Apple has used proprietary interfaces for their SSDs, while most other consumer-orientated computers have used SATA or M.2. Recently, Seagate developed Multi Actuator technology for their advanced hard disks (see Fig.34). The actuator is the part that moves the hard drive heads. Until now, hard drives had only one actuator to move 2023 2024 2025 22TB 30TB+ 40TB? 0.39L 0.39L 0.39L 0.7kg 0.7kg 0.7kg 8.5ms 8.5ms 8.5ms 2.1¢ ~1.5¢ ~1.2¢ 195Gb/cm2 290Gb/cm2 >350Gb/cm2 2,500,000 2,500,000 ~2,500,000 siliconchip.com.au Australia's electronics magazine An old hard drive legend Massive old ‘washing machine’ hard drives could ‘walk’ around the floor in response to certain head access patterns. There is an unverified legend that once such a drive walked so far that it blocked the only door to the room, and a hole had to be cut in the wall to gain access! March 2024  17 Fig.34: the Seagate Multi Actuator is two independent sets of heads that can double data throughput. all heads simultaneously. That means that all the heads are always over the same track. Seagate uses two actuators so half of the heads can move independently and simultaneously with the other heads, increasing the data throughput. Effectively, the drive acts like two separate drives in one case. You can see how it works in the video at https://i. imgur.com/uZaizwd.mp4 Another advanced technology developed by Seagate is HAMR, or heat-assisted magnetic recording. To make higher data density disks with smaller magnetic domains, materials that are harder to magnetise (and retain magnetisation better) are needed so that small areas remain stable. The heat from a laser in the head assists the magnetisation process. A dot is heated to 450°C, magnetised and then returned to room temperature in one nanosecond! Another recent development is using helium as the gas inside a hard drive. The idea was conceived in the 1970s, but after numerous failed attempts, it was thought to be impossible due to problems with containing the helium. Research resumed in 2009 at Hitachi, which was acquired by Western Digital (WD) in 2013, and Seagate bought WD in 2014. WD now makes about one million helium-filled drives per month – see Fig.35. Seagate also sells them under their own brands, such as Exos and IronWolf Pro. In fact, many hard drives with capacities of at least 8TB sold in the last few years are helium-filled. Helium has around 1/7th the density of air, with much lower viscosity, resulting in much less turbulence and friction inside the drive. That means a much cooler running drive, lower power consumption and less noise. This lesser friction means the drive’s platters can be thinner, allowing for up 18 Silicon Chip to 10 platters instead of 6 in the same size, according to WD. More heads can also be used. Also, since helium-filled drives are completely sealed, atmospheric contaminants can’t enter through the breather port that exists in air-filled drives. Anyone who has worked with helium knows it is notoriously hard to contain, and it will eventually leak out. However, WD says that the helium will remain through the operational lifetime of the drive. Finding a way to hermetically seal the hard drive to keep the helium in was a major challenge during their development. The famous first image of a black hole, or more correctly, its surrounds, was made with the assistance of WD helium-filled hard drives, as it required the acquisition and analysis of 4.5 petabytes of data. Perpendicular recording is a process by which magnetic domains are written in a vertical manner rather than a longitudinal manner. This allows three times the data density of longitudinal writing. Shingled magnetic recording (SMR) is a hard drive technology where data tracks are written slightly overlapping each other, like roof shingles, rather than with gaps between each row, as in earlier drives. This allows higher track density. However, this strategy requires extensive management of the data by firmware within the drive, as whenever a single bit of data needs to be changed, the entire ‘shingle’ has to be rewritten in order due to the overlaps. As far as the computer’s operating system is concerned, though, it appears as a normal drive. SMR drives generally have a high data throughput and reasonable seek performance. Still, the performance will plummet dramatically if many ‘random writes’ are performed without giving the drive time to ‘rest’ (during which it reorganises data and rewrites the shingles). That resulted in WD being sued by customers when they sold SMR hard drives without labelling them as such, as they are unsuitable for certain workloads (siliconchip.au/link/absa). They are mainly used as ‘online backups’ or video recording; applications that involve writing data in large batches. Modern hard drives can be mounted and used in any orientation, including upside-down or sideways, as long as cooling is adequate. That was not necessarily the case for earlier PC hard drives, before ‘flying heads’, as it could affect the head gap and cause data previously written to become unreadable. Then again, with the early washing-machine-sized hard drives, you didn’t have much choice in orientation! The Internet Archive (https:// archive.org/) is a vast free library of information and uses many hard Fig.35: banks of Western Digital HelioSeal hard drives in a data centre. Source: https://documents.westerndigital.com/content/dam/doc-library/en_us/assets/ public/western-digital/collateral/brochure/brochure-helioseal-technology.pdf Australia's electronics magazine siliconchip.com.au disks. As of December 2021, they had 28,000 spinning disks spread across 745 nodes in four data centres. The Wayback Machine internet archive contains 57 petabytes; the book, music and video collections contain 42 petabytes; the amount of unique data is 99 petabytes, and the total storage used is 212 petabytes. Data is held in storage units called petaboxes (https://w.wiki/8Xxe), with 1.4 petabytes per rack. One petabyte is one million gigabytes or 1000 terabytes. Miniature hard drives Kittyhawk was a miniature hard disk introduced by Hewlett Packard in 1992, with a 1.3in (3.3cm) form factor and a capacity of 20MB (later, 40MB). It was discontinued in 1994, being a commercial failure. Microdrive was a miniature 1in (25mm) hard drive format produced by IBM and Hitachi and designed to fit into CompactFlash Type II slots – see Fig.36. They were introduced in 1999 and last produced around 2007. They were used in devices such as cameras, printers, iPods and anywhere else a flash memory card was useful. They provided a higher capacity than flash memory at the time and at a lower cost. In addition to IBM (170MB to 16GB) and Hitachi (512MB to 8GB), the technology was used by the Seagate ST1 (2.5GB to 12GB), GS Magicstor (2.2GB to 6GB), Sony (2GB to 8GB), Western Digital (6GB), Cornice (2GB to 8GB) and Toshiba (2GB and 4GB). Flash memory Flash memory is a form of erasable, nonvolatile memory, usually in the form of NOR flash or NAND flash. Fig.37: both NAND and NOR flash store data using floating-gate Mosfets; the difference is in how the memory cells are addressed. NAND flash has higher density & faster write speeds, while NOR is more reliable and can be read faster. NOR and NAND are types of logic that are formed by the structure of the flash blocks. The NOR function is OR with the output inverted, while NAND is AND with the output inverted. The different layouts are shown in Fig.37. Whichever type of logic is used, the fundamental design is based on floating gate Mosfet memory cells. A charge is kept within highly insulating materials, and the logic inputs are only capacitively coupled to it, so the charge, and thus the memory bit it represents, can be maintained for a very long time, at least ten years (probably much more) with current technology. Toshiba invented Flash memory in 1980 and marketed it from 1987, Fig.36: one of the later Microdrives; this one was produced by Seagate and stored 5GB. It’s the same size as the earlier IBM models, though. The 50 Euro cent coin is the same size as our $1 coin. siliconchip.com.au Australia's electronics magazine although Dawon Kahng and Simon Min Sze invented the floating gate Mosfet at Bell Labs much earlier, in 1967. NOR flash is optimised for random access; individual memory cells can be accessed. NAND flash is optimised for high-density storage and forgoes random data access. Because of its architecture, individual memory cells cannot be accessed, as with NOR. They have to be read and written a block at a time. Because NAND offers a higher data density, it is used in devices like memory cards, USB drives and SSDs that require a large storage capacity. NOR has a lower data density with larger cell sizes, is less prone to data corruption and is used in applications such as code execution in medical devices or mobile phones where high capacity is unnecessary, but reliability is. Because individual cells can be addressed, NOR flash enables fast read times but relatively slow write and erase times due to the large cell size. NAND flash reads are slower because whole data blocks must be read in one go. However, writing and erasing is quicker than with NOR. NAND flash has a lower cost for a given capacity. Flash memory is slower than static RAM or ROM memory. In 2007, Toshiba introduced three-­ dimensional NAND architectures, March 2024  19 Fig.38: the basic structure of 3D NAND flash memory. SGD = drain-end select gate, SGS = select gate line, WL = word line, BL = bit line. Source: Toshiba Corporation. such as the generic 3D architecture shown in Fig.38. 3D NAND flash allows a much greater capacity in one package. Flash memory has only a finite, although high, number of write cycles as it ‘wears out’. Strategies must be implemented to keep this wear even across all memory cells by ‘wear levelling’ and other techniques within the drive, to delay the inevitable wearing-­ out process as much as possible. With wear levelling, the number of writes to each block is tracked, and when there is a choice, the next block to be written is the one with the lowest number of write cycles so far. To allow this, the controller performs logical-­ to-physical block mapping, allowing it to rearrange currently unallocated blocks at will. Memory cards Flash memory cards are usually based on the flash memory technology described above. There have been many variations over the years, some of which are shown in Fig.39. Table 2 shows how flash chip capacity, cost and speed have changed over time. PC Card (previously PCMCIA, Personal Computer Memory Card International Association) was introduced in 1990 and renamed in 1995. The format was initially designed for memory but later adapted to many other uses, as a convenient way to add peripherals to portable computers. It was superseded in 2003 and replaced with ExpressCard, which became obsolete in 2018 (it was never popular). Linear Flash cards are a PC Card format and are obsolete, but they are still used in various devices and still available for purchase, presumably for military and industrial applications. SRAM is another type of PC Card format memory card that requires a battery to maintain the memory. CompactFlash (CF) is a flash memory card format introduced by SanDisk in 1994. They were initially based on NOR memory but later switched to NAND. The low density of NOR flash is one reason the cards are relatively large. The other reason is that they were designed to be compatible with PCMCIA, using a 50-pin subset of the 68-pin PCMCIA interface. The original CF cards had capacities of 2-15MB at speeds of up to 8.3MB/s (but usually much slower), although the original specification supported capacities up to 128GB. Miniature Card (37 × 45 × 3.5mm) was developed by Intel and first promoted in 1995. It was backed by AMD, Fujitsu and Sharp. It is obsolete, having been available from around 1997. The maximum capacity was 16MB, and it was used in some digital cameras, such as the first HP PhotoSmart and the Intel 971 PC camera kit. SmartMedia Card was introduced by Toshiba in 1995 and discontinued in the early 2000s. One of the intentions of the card was to replace the 3.5in floppy disk; there was even an adaptor to insert them into a 3.5in drive bay. Cards could be written to by a camera, then read in a computer’s floppy drive via an adaptor. Cards from 2MB to 128MB were released. There was no in-built controller chip and therefore no wear levelling to extend the card’s life, so cards often became corrupted or unreadable. It was a popular media in digital cameras at the time, especially with Fuji­ film and Olympus. The Serial Flash Module was introduced in 1996 and discontinued in 2003. Capacities were from 128kB to 4MB; it was renamed to MediaStik in the early 2000s. MultiMediaCard (MMC) was introduced in 1997 by SanDisk and Siemens. SD cards (described below) evolved from MMC; some devices support both SD cards and MMCs. However, MMCs are thinner at 1.4mm compared to SD cards, which are 2.1mm thick, so MMC cards may fit into an SD card slot but not necessarily vice versa. MMC has been released in several varieties and form factors such as RS-MMC, DV-MMC, MMCplus, MMCmobile, MMCmicro, MiCard and eMMC. MMC has lost popularity now, but eMMC, an embedded, non-removable type of memory, is still used for storage in many phones and other devices. Fig.39: a selection of flash memory cards. Source: https://w. wiki/8XxK (CC BY-SA 3.0). 20 Silicon Chip Australia's electronics magazine siliconchip.com.au However, since 2016, when Universal Flash Storage (UFS, see below) was released, it has come to dominate that market. One advantage of MMC over SD is its low cost, and eMMC is cheaper than other forms of embedded storage in phones, such as an NVMe solid-­ state drive. Memory Stick was a proprietary flash memory technology launched by Sony in 1998. Its original format ceased to be available in 2007. Memory Stick PRO-HG Duo HX was released in 2011 and is still available in sizes up to 128GB. They appear to be no longer under active development. There are adaptors to use microSD cards in some devices that require Memory Stick Pro Duo cards (see siliconchip.au/link/abry), but if you are considering buying one, do some research as they have limitations. Sony now makes its own SD cards. USB Flash Drives (‘thumb drives’) are one of the most ubiquitous portable storage devices, often attached to key rings or neck lanyards. These drives originated in 1999 when Amir Ban, Dov Moran and Oron Ogdan of M-­ Systems in Israel filed a patent application entitled “Architecture for a Universal Serial Bus-Based PC Flash Disk” and subsequently were awarded US Patent 6,148,354. Those people are generally recognised as the inventors; there are other claimants, but they did not file for patents. A USB flash drive contains a USB controller and one or more flash memory chips – see Fig.40. SD (Secure Digital) cards are a form of flash memory used (originally) in the form of a postage stamp size module, although much smaller formats are now available. They are primarily used in portable devices like phones and cameras. The format was introduced in 1999 by Panasonic, SanDisk and Toshiba as an improved version of MMC cards. The standards are governed by the SD Association (www.sdcard.org). Formats smaller than the original include miniSD (no longer produced) and microSD (shown opposite). Standard SD cards had a capacity of up to 2GB. SDHC cards were introduced in 2006, ranging from 2GB to 32GB. SDXC cards were introduced in 2010 and have capacities of 32GB to 2TB. We published an article primarily on SD cards (but that also siliconchip.com.au Table 2: flash memory chip evolution since 1990 (per chip) 1990 1995 2000 2005 2010 2015 2020 2023 2MB 16MB 2GB 64GB 256GB 1TB 2TB Read/write 500kB/s speed 2MB/s 5MB/s 25MB/s 100MB/s 250MB/s 1GB/s 2GB/s US$/chip $300.00 $40.00 $20.00 $40.00 $40.00 $100.00 $100.00 $60.00 $20,000 $1,200.00 $20.00 $0.62 $0.40 $0.10 $0.03 Capacity 512kB US$/GB $600,000 mentioned other flash memory cards) in the July 2013 issue (siliconchip.au/ Article/3935). In 2019, SDUC cards were introduced with theoretical capacities of up to 128TB. There are also various speed categories for SD cards, such as Default, High Speed, Ultra High Speed (UHS), UHS-1, UHS-II (with extra pins), UHS-III (also with extra pins) and SD Express. SD Express cards have extra pins to support a PCIe lane and the NVM Express memory access protocol. Some SD cards even have integrated WiFi to automatically offload data wirelessly. The xD-Picture Card was introduced by Fujifilm, Kodak and Olympus in 2002 and discontinued around 2009. The largest capacity released was 2GB. These cards have no ‘flash translation layer’ to emulate a hard disk; the NAND flash hardware is (more or less) accessed directly. It was derived from the SmartMedia card and, like that, has no wear-levelling controller. P2 was a professional memory card format introduced by Panasonic in 2004, available in capacities up to 64GB. They are still listed on the Panasonic website (siliconchip.au/link/ abs6) and are described as having “four SD cards packaged into one” (device). They are packaged into a PC Card (formerly PCMCIA) and were replaced by the compatible MicroP2 (based on SDXC/SDHC). SxS is a flash memory storage card developed by Sony and SanDisk and first announced in 2007, followed by SxS Pro cards in 2011. It is designed for professional video cameras, with an emphasis on high performance and reliability. It is compatible with the ExpressCard/34 interface or USB via an adaptor. Cards from 32GB to 240GB are available from Sony’s website. CFast flash memory cards were introduced in 2009. The format is supported by relatively few cameras; mostly high-end professional cinema cameras from Arri, Atomos, Blackmagic Design and Canon. It is used in still cameras such as the Canon EOS-1D X Mark II and Hasselblad H6D-100C. We have seen CFast 2.0 cards up to 1TB capacity. XQD flash memory cards were developed for high-definition camcorders and cameras. The format was developed by Nikon, SanDisk and Sony and was introduced to the market in 2012. Currently, the cards are available with a capacity of up to 2TB. XQD cards are still available but have been succeeded by CFexpress, which Fig.40: an old 64MB USB flash drive removed from its case. The key components are 1) USB connector, 2) controller, 3) test connectors, 4) NAND flash memory, 5) crystal, 6) LED, 7) writeprotect switch, and 8) space for a second flash chip. Source: https://w. wiki/8XxJ (GNU FDL). Australia's electronics magazine March 2024  21 Fig.41: a comparison of the read/write schemes for eMMC and UFS; LVDS is low-voltage differential signalling. UFS cards are faster because reads and writes can occur simultaneously, and there is command queuing. Fig.42: a comparison of how the electrical interfaces work with SD and UFS cards. is backwards compatible with XQD (for Type B cards). AXS memory cards are a proprietary format for Sony high-resolution digital F55 and F5 cinematography cameras, with a capacity of up to 1TB. They were introduced around 2012. It is not a standard, but we included it in case you wondered what cards are used for certain cinema cameras. Sony SRMemory cards are related to AXS, for use with the Sony SR-R1 portable recorder for HD-SDI (High-­ Definition Serial Digital Interface) cameras. CFexpress is a format for flash memory cards launched by the CompactFlash Association in 2017. They are available in types A, B and C. Type B slots will accept XQD cards. We have seen CFexpress cards with capacities of up to 4TB. Universal Flash Storage (UFS) is a flash storage system designed to be faster, more reliable and use less power than eMMC for internal storage and SD cards for external storage in devices such as cameras, phones and others – see Fig.41. It is intended to replace those two technologies. UFS achieves higher speeds for internal memory than eMMC because UFS has dedicated channels for reading and writing, so reading and writing can occur simultaneously, unlike with eMMC. Also, UFS has command queuing to organise read and write commands in the most efficient manner. According to Samsung, a UFS card is up to 70 times faster than an SD card. UFS memory cards have been designed in a similar form factor to SD cards so that a single slot can accept either a microSD card or a UFS card. It achieves that by placing the contacts for both devices in unique locations, except for the shared power pins; see Fig.42. A UFS card is faster than an SD card in external memory card applications because it has a high-speed serial interface with separate data channels for transmitting and receiving, enabling simultaneous operation. UHS-II and UHS-III SD cards used a similar approach to boost transfer rates, but the UFS serial interface is still faster – see Fig.43. Solid-state drives (SSDs) SSDs are gradually replacing hard disks in applications where a high capacity is not critical, like the boot 22 Silicon Chip Australia's electronics magazine siliconchip.com.au drives of portable and desktop computers. SSDs typically use flash memory for storage. Advantages over traditional hard disks include greater robustness (at least in theory, due to a lack of moving parts), higher speeds, especially for ‘random’ I/O, and silent operation. Most SSDs use NAND flash memory of several possible design types. Flash memory may contain 1, 2, 3, 4 or 5 bits of data per cell. These cells are known as Single-Level Cells (SLC), Double or Multi-Level Cells (DLC/MLC), Triple-Level Cells (TLC), Quad-Level Cells (QLC) or Penta-Level Cells (PLC). As more bits are added per memory cell, there are trade-offs of performance, endurance and reliability. SLCs are the most reliable and fastest, but the most expensive per unit of capacity, so they are suitable for enterprise operations with intensive write operations. The upcoming PLCs offer the lowest cost and highest data density but with the least durability, so they are suitable for large data applications with low-­intensity workloads. SSDs may contain a mix of technologies, eg, some SLC cells for frequently accessed data and many MLC, TLC, QLC or PLC cells for long-term storage. Multi-level cell flash can even ‘emulate’ SLC for faster read/write speeds but lower density, providing a ‘cache’ without needing actual SLC flash. Given the capacities of SSDs and the fact that they are expected to store data long-term, good wear-levelling algorithms are essential. Also relevant to SSDs are the sections above on flash memory, wear-­ levelling, 3D flash technology and hard disk interfaces. While flashbased SSDs can use the same interfaces as mechanical hard disks, the NVMe/M.2 and mSATA interfaces are almost exclusively used for SSDs. Such devices are shown in Fig.44. NVM Express (NVMe or Nonvolatile Memory Host Controller Interface Specification [NVMHCIS]) is an open standard and a logical interface protocol for nonvolatile storage devices, usually attached via PCI Express bus (see https://nvmexpress.org/). It was implemented because existing interfaces like SATA were not fast enough for the latest SSDs. It exploits the parallelism possible in solid-state memory devices and the fact that the SSDs are smaller and thus can be kept closer to the motherboard. This siliconchip.com.au My experience with the longevity of SD cards I had some old SD cards, which I had used in a camera, plus some old USB flash drives. Some had not been used for 10 or 20 years. When I went to read them, I had no problems, suggesting that data should last at least that long. However, it is always wise to have backups and also to “refresh” the cards by putting them in a reader every so often and allowing the card’s internal firmware to correct any fixable defects, plus replace any lost charge on the floating-gate Mosfet transistor used to store bits of data. Note that there’s no guarantee that modern flash memory has the same longevity; it will likely have smaller cell sizes and thus possibly won’t retain data for as long as older flash chips. Fig.43: the physical differences between UFS and microSD cards. They both fit in a combination reader. Source: https://semiconductor.samsung.com/newsevents/tech-blog/ufs-solutions-high-performance-storage-solution/ Australia's electronics magazine March 2024  23 Links and further reading ● ● ● ● ● ● ● ● ● ● Practical applications of the punched card: siliconchip.au/link/abs0 Appletons’ Cyclopaedia of Applied Mechanics: siliconchip.au/link/abs1 The IBM Diskette General Information Manual: siliconchip.au/link/abs2 The IBM 1311 Disk Storage Drive manual: siliconchip.au/link/abs3 IBM 1360 Photo-Digital Storage System manual: siliconchip.au/link/abrv Introduction to IBM Direct Access Storage Device: siliconchip.au/link/ abs4 “1951-1968 Early Computer Magnetic Tape Units”: https://youtu.be/ lEYyZSlQEdg “Debugging the 1959 IBM 729 Vacuum Column Tape Drive”: https://youtu. be/7Lh4CMz_Z6M “Making a bootable OS/8 DecTape for the PDP8/m”: https://youtu.be/ tOWt7LIOVJs “DECTAPE II, TU58, & TEAC MR-30 Transport”: https://youtu.be/jo4qfVl-Y-o specification was introduced in 2011 and last updated in April 2022. Larger devices can use more than the four PCI Express lanes provided by an M.2 connector, such as the large SSD shown in Fig.45. Bit rot One important drawback of the MLC/TLC/QLC/PLC cell structure that is not widely known but that we should mention is the performance degradation over time. Just after data has been written to a flash cell, its voltage should be well within the defined thresholds, so reading it back will be very fast. However, over time (months or years), the voltage will drift due to tiny leakage currents. If the voltage drifts far enough, it could cross one of the boundaries and the data will become corrupted (unless the SSD has built-in error checking and correction; we expect many would). However, even if the voltage doesn’t drift far enough to cause data loss, it can still slow down reading significantly. That’s because the high-speed amplifiers/comparators that read data out of the flash are noisy and imprecise, so they only work properly when the voltage is within a narrow band. Once it drifts outside that band, a slower and more precise method has to be used to determine the stored data. That means that the read speed of an SSD can drop dramatically, from gigabytes per second to just a few megabytes per second, if the particular file hasn’t been touched after a few months or years. In our experience, it isn’t quite so dramatic, dropping to maybe 50MB/s, but that’s still far shy of the expected read performance of an SSD. This seems to affect many makes and models of SSDs and the only complete solution is to periodically (eg, every few months) perform a complete ‘refresh’ of the drive by reading and then rewriting all data. However, most drives and operating systems don’t (yet) do that automatically. There is software available to do it. In our experience, some SSDs will automatically refresh such files when read. So it’s only slow the first time you access a file that was written a while ago. Not all do that, though, and you may be forced to rewrite an older file to fix the slowness. Ideally, the SSD will periodically scan its own data, find blocks that haven’t been touched in a while and refresh them automatically. However, that does not yet seem to be a common feature of SSD controllers. Maybe it will be one day. Exabyte-scale storage CERN (Conseil Européen pour la Recherche Nucléaire or European Council for Nuclear Research) in Switzerland now has a storage capacity of one exabyte of data (or one million terabytes or 1000 petabytes) to store data from experiments at the world’s largest particle accelerator. The data is stored in 111,000 devices, primarily hard disks with an increasing number of SSDs; see Fig.46. Long-term archival storage Spacecraft Voyager 1 and 2 carry a Golden Record, a 12in (30cm) goldplated copper disc containing pictures and sounds of the Earth. It was the first time a library was taken into space. We described the record in our article on Voyager (December 2018; siliconchip. au/Article/11329). The Beresheet Lunar Library was the second attempt to send a library into space. The library comprised data stored in DNA and on nickel disks. The Fig.44 (left): an mSATA SSD is on the left, while an M.2 NVMe SSD is on the right. Source: https://w.wiki/8XxM (CC BY-SA 4.0). Fig.45: an Intel solid-state drive for a desktop computer or server that plugs into a PCI Express 8x slot. M.2 NVMe drives use a similar interface but with fewer lanes on a smaller connector. Source: https://w.wiki/8XxL (CC BY-SA 4.0). 24 Silicon Chip Australia's electronics magazine siliconchip.com.au contents included a 30 million page archive of ‘human history and civilisation’ on a 100mm nanotechnology-­ fabricated device similar to a DVD. It contained 25 discs, each 40 microns thick, see Figs.51 & 52. The first four discs were analog and contained 60,000 images etched from low resolution to increasingly high levels of information up to the nanoscale, made with optical nanolithography. The analog front cover has information visible to the naked eye, plus smaller images and holographic logos. The discs also carry information on many human languages. In total, all the discs carried around 200GB of digitally compressed content. Even though Beresheet crashlanded on the moon, it is thought that the contents of its library remained intact. We had a detailed article on the landing attempt in the November 2018 issue (siliconchip.au/Article/11296). The Arch Mission Foundation (www.archmission.org) is a non-profit organisation aiming to preserve all human knowledge by building data archives. This is so that, in the event of a calamity, it would be much easier to rebuild civilisation (if anyone survives). Lunar Library 1 in the Beresheet Lunar Library was one of their projects. Fig.46: a few of the 111,000 devices that make up one exabyte of storage at CERN. Source: https://home.cern/news/news/computing/exabyte-disk-storage-cern The future of data storage Storage technologies are still evolvingl; the future of data storage technologies includes: In hybrid cloud storage, less frequently accessed data is stored offsite ‘in the cloud’ and more frequently accessed data is stored on the premises. Multi-cloud storage is where multiple cloud storage vendors are utilised to avoid dependency and the risk of being with just one provider. Quantum data storage uses quantum atomic properties such as superposition and entanglement to potentially encrypt and store large amounts of data (see Fig.47). Information is kept in qubits instead of being represented as 0 or 1 bits like in regular memory. A qubit is 0 and 1 simultaneously, vastly increasing the capability of such memory and computer systems. Just 100 qubits could hold more information than all of the world’s hard disks, according to Doug Finke of the Quantum Computing Report. However, such a system is highly susceptible to ‘decoherence’, where siliconchip.com.au Fig.47: a circuit model for Quantum RAM. Original source: https://ncatlab.org/ nlab/show/QRAM Fig.48: the growth of hard drive (HDD), flash and optical data storage (ODS) capacity from 1980 to 2014, with projections to the present. Source: Figure 8 from “Optical storage arrays: A perspective for future big data storage” – siliconchip.au/ link/abs8 (CC BY-NC-ND 3.0). Australia's electronics magazine March 2024  25 would have to consider the size of the coding and decoding equipment in a DNA data storage system. It has been estimated that 1g of DNA molecules could store about 215 petabytes of data (a petabyte is one million gigabytes). The entirety of Wikipedia (16GB in 2019) was turned into synthetic DNA, as described at siliconchip.au/ link/abrz The Beresheet Lunar Library mentioned earlier also contained 10,000 images and 20 books encoded in DNA. Fig.49: two ways data can be stored in DNA, either by sequencing or structure. Source: https://pubs.acs.org/doi/10.1021/acsnano.2c06748 (CC-BY 4.0). the information is destroyed; a significant problem, to say the least! Such memory is called Quantum RAM or qRAM, the quantum equivalent of classic RAM. Also see our article on Quantum Computing in the March 2016 issue (siliconchip.au/Article/9845). Edge storage is where data is stored and processed close to where it is generated rather than, say, in the cloud. The maximum size of hard disks is expected to increase to 100TB by 2025, according to the Storage Technology Consortium (https://idema.org/) – see Fig.48. That figure is from 2014, and the projections to present have already been exceeded. For example, hard disks were projected to have a 1.5TB technical limitation, but that has been far exceeded, and 28TB drives are now available (using shingled magnetic recording and helium filling). A Seagate 32TB hard disk using HAMR (heat-assisted magnetic recording technology) is said to be in production. It should be available to purchase by the time this article is published. Tom’s Hardware claims 40TB+ drives will be on the market by 2025. We doubt that 100TB will be reached by 2025, but it likely will be eventually. Holographic data storage is a future scheme where data is stored in optical media as an interference pattern. According to one estimate, holographic memory has the potential to store 1TB of data in the size of a sugar cube. However, bear in mind that 1TB SD cards are available and occupy less volume than that. For more, see the video “How does holographic storage work?” at https:// youtu.be/4EADwGV5Gv8 DNA Storage (Figs.49 & 50) uses the double-helix-shaped molecule that encodes genetic instructions for virtually all living organisms. Information is encoded as combinations of four so-called nucleobases: cytosine (C), guanine (G), adenine (A) and thymine (T). Information density is exceptionally high since information is stored at the molecular level. DNA is relatively stable (good news for us!) and can last hundreds or thousands of years under the right circumstances. Disadvantages are that reading and writing processes are slow and can be error-prone. To encode DNA with data, bytes or tokens are first converted to a corresponding unique DNA sequence, such as shown in Table 3. The density of DNA storage is hard to give a precise figure for because you 5D optical storage 5D optical storage has been researched as part of Microsoft Project Silica (see Figs.53 & 54). Data is written by the use of a femtosecond laser focused inside a piece of quartz glass, where it causes damage and forms a voxel (volumetric pixel) located within a three-dimensional (X/Y/Z) space also with properties of volume and orientation, which add extra data apart from the spatial position. That leads to the prefix ‘five dimensions’ or ‘5D’, even though it is physically only 3D, as each voxel has five properties. Data is read by a microscope-like device. The technology is read-only (or at least WORM [write once read many]) and is intended for archival storage. Data can be stored for thousands of years, and it is resistant to damage and degradation. Microsoft suggests a capacity of 7TB in a glass platter the size of a DVD. For more information, see the video “Project Silica - Storing Data in Glass” at https://youtu.be/6CzHsibqpIs Keeping data long-term It is important to make sure data in obsolete formats are migrated to more modern formats. In 1985, there was a rumour that US Census data from the 1960s had been lost. The claim was that “The Fig.50: the six steps of DNA data storage. Source: https://pubs.acs.org/doi/10.1021/acsnano.2c06748 (CC-BY 4.0). 26 Silicon Chip Australia's electronics magazine siliconchip.com.au Fig.51: the front cover disc of The Lunar Library on the Beresheet lunar lander. Fig.52: a detail of one of the images of the front cover of The Lunar Library. siliconchip.com.au Australia's electronics magazine March 2024  27 Table 3: proposed ASCII to DNA encoding scheme ‘ ‘ ACAT <at> CCAC „ TCCG ! AGGT A TACT a GAGC “ AAAG B TCCT b GTGC # AGAC C TACG c GACG $ AAGC D TGCC d GTAA % AACT E TCTA e GTAC & AGAA F TAGT f GCCT ‘ AATC G TTAA g GCTA ( ATTG H TGGC h GAGT ) AATT I TGTT i GATG * AATG J TTCC j GATT + AAGA K TACT k GGGC , AGAG L TATG l GTTG - AAGC M TAGT m GTGA . ACAC N TGTC n GACT / ACGT O TATT o GCCG 0 CAAA P TTCA p GACA 1 CACC Q TTTA q GACT 2 CCGT R TAGA r GGAT 3 CGAG S TGAG s GGTG 4 CCTT T TAAA t GCTT 5 CCGT U TGAC u GACC 6 CTGT V TGAG v GACT 7 CTCT W TAAC w GCCC 8 CCGT X TCCT x GATC 9 CTCA Y TGAA y GTCG : CTAG Z TAAG z GTGA ; CCGC [ TCAT { GGCT < CACA \ TAAG | GGTG = CATA ] TCCA } GAAC > CTAC ^ TGTT ~ GATG ? CCAG _ TCCG DEL GAGT Fig.53: this 75 × 75 × 2mm piece of glass from Project Silica contains the 1978 Superman movie. It was produced in 2019 and stored 75.6GB. New versions store much more data. Source: https://news.microsoft.com/source/features/ innovation/ignite-project-silica-superman/ Source: “Design and Implementation of a New DNA Based Stream Cipher Algorithm using Python” – siliconchip.au/link/abs9 Fig.54: how a microscope can read Project Silica quartz glass with a green laser. The top view (left circle) shows vertical columns of voxels. The colours represent the different volumes and orientations of each voxel, and the side view (right circle) shows the layers of the voxels, each with a different size and orientation. Source: https://youtu.be/6CzHsibqpIs 1960 Census, for example, was written on tapes for the Univac I, a machine that has been obsolete for more than two decades. Its obsolescence caused much of the census data to be lost.” Fortunately, contrary to popular belief, the data was migrated in that case. Quoting from siliconchip.au/ link/abs7: By 1979 the Census Bureau reported that they had successfully completed copying 640 of the 642 II-A tapes onto 178 industry-compatible tapes. ... a small volume of records from the 1960 census was lost. This occurred because of inadequate inventory control and because of the physical deterioration of a minuscule number of records, not technological obsolescence. From what we have described in these two articles, you can see the huge variety of secondary storage used in the past that has become obsolete while new types continue to be developed. Thus, important data must frequently be migrated from outdated media to new media to preserve it. You must also be aware of the possibility of ‘bit rot’, where data on old media such as floppy disks becomes corrupt over time, a problem the author (and Silicon Chip) has experienced. 28 Silicon Chip Australia's electronics magazine This is especially a problem for modern SSDs; we understand that, in some cases, simply leaving them powered off for a few months can lead to data loss. Most SSDs are not intended to be used for archiving purposes, but rather actively written and read daily or near-daily. Mechanical hard disks also require frequent (eg, monthly) ‘scrubbing’ where the entire disk is read and then rewritten for reliable long-term data storage. That’s because the magnetic domains are so small that untouched areas can eventually lose enough magnetisation to become unreadable. SC siliconchip.com.au