Skip to content

Hesela / Chronology

Storage history

From punched records to networked flash.
A history of media, interfaces and the systems around them.

1890–202430 milestonesAbout the evidence

1890–1959

Records become machine-readable

From punched records to magnetic tape and direct-access disks.

  1. 01

    Punched cards count the US census

    Herman Hollerith's machine-readable punched cards were used for the 1890 US Census. Earlier punched-card uses included control of textile machinery.

    Why it matters Machine-readable records connected information on paper to electrical counting and assisted sorting.

    History & engineering

    Historical context

    The Census Office needed to turn a growing volume of handwritten returns into usable statistics. Hollerith won an 1888 processing competition before his system was selected for the 1890 census. Clerks still had to transfer each person's answers from the census schedule onto a punched card.

    How it works

    A pantograph helped position the holes. In the reader, spring-loaded pins passed through those holes into mercury contacts, closing electrical circuits that advanced counting dials. An adjacent sorting table indicated where the operator should file the card. The system combined paper records, electrical sensing and human handling.

    Limits & interpretation

    This was electromechanical data processing, not a stored-program computer or an automatic scan of the original forms. Encoding errors and physical handling still mattered. The milestone is the census deployment, not the invention of all punched media; the familiar IBM 80-column format arrived much later, in 1928.

    Sources (2) Documented milestone
  2. 02

    IBM introduces the 80-column card

    IBM introduced an 80-column punched card in 1928. Rectangular holes increased the data that fitted on a card; this was a format change, not the invention of punched media.

    Why it matters The physical layout of a record became part of the data-processing system.

    History & engineering

    Historical context

    The challenge was to increase the information held by an established physical card without enlarging it. IBM evaluated competing designs before choosing smaller rectangular punches for its 1928 format, making media geometry a central part of the processing system.

    How it works

    The 80 columns provided positions for encoded characters. IBM's history distinguishes the initial numeric format from its 1930 revision with 12 rows. Punching, reading and interpreting the card depended on matching equipment and coding conventions, not simply on counting holes.

    Limits & interpretation

    More columns did not make old machines automatically compatible. IBM's account explicitly connects the chosen design to its own equipment. This illustrates an enduring storage trade-off: a denser format can increase usefulness while also tying records to particular readers and an ecosystem.

    Sources (1) Documented milestone
  3. 03

    Vacuum columns make tape practical at speed

    IBM introduced vacuum-column tape handling for the IBM 701 in 1953. Buffered loops allowed the tape at the head to start and stop without accelerating a full reel as abruptly.

    Why it matters Mechanical buffering helped reconcile fast data access with fragile moving media.

    History & engineering

    Historical context

    Early computer tape handling had to reconcile rapid starts and stops near the recording head with the inertia of much larger reels. Simply pulling harder risked stretching or breaking the thin medium. IBM's vacuum-column work addressed this mechanical bottleneck.

    How it works

    A partial vacuum held a loop of tape in a column. The loop acted as a buffer between reel motion and the short section passing the head, so the whole reel did not need to follow every abrupt change at the recording point.

    Limits & interpretation

    The buffer improved transport control; it did not turn tape into random-access storage. A distant record still required moving the medium. The lesson is that storage performance depends on controlling the physical path as well as encoding the bits.

    Sources (1) Documented milestone
  4. 04

    IBM 350 RAMAC — first commercial HDD

    IBM introduced RAMAC with the 350 disk storage unit: five million alphanumeric characters on fifty 24-inch disks. Character data, parity and recording overhead are distinct from modern byte capacity.

    Why it matters Random access let a machine retrieve a record without reading a tape from the beginning.

    History & engineering

    Historical context

    Business records did not always arrive in the order in which a company needed to retrieve them. RAMAC addressed that mismatch with an addressable disk file, allowing a machine to reach a requested record without working through all preceding tape records.

    How it works

    A moving arm positioned magnetic heads at the required disk surface and track while the disk stack rotated. Reading or updating a record therefore combined head positioning with waiting for the relevant recording location to pass the head.

    Limits & interpretation

    Random access meant a different access pattern, not the absence of mechanical delay. The machine's five-million-character rating also belongs to its original encoding system; treating that figure as five million modern eight-bit payload bytes would obscure the distinction between characters and recorded overhead.

    Sources (3) Documented milestone

1960–1979

Storage becomes more portable

Removable media and new head designs reshape the drive.

  1. 05

    The floppy disk reaches IBM systems

    IBM began selling floppy disk drives in 1971, following work on an inexpensive way to load instructions into mainframe systems. The drive and disk received US patents in 1972.

    Why it matters Removable magnetic media offered a practical route for moving programs between machines.

    History & engineering

    Historical context

    IBM's initial problem was distributing instructions to mainframe equipment economically, rather than giving personal-computer users a universal file-sharing format. The later association with desktop software should not be projected backward onto this first product generation.

    How it works

    The medium was a flexible magnetic-coated disk that rotated inside a protective envelope. A head reached the recording surface through an opening. IBM's design history describes how the protective enclosure helped address dust and handling concerns.

    Limits & interpretation

    Removability made the medium convenient to transport, but exchanging the disk did not by itself establish format compatibility. Physical size, recording layout and the receiving system still mattered. The 1971 product introduction and the 1972 patents are separate events.

    Sources (1) Documented milestone
  2. 06

    Winchester architecture

    IBM's 3340 paired low-mass heads and lubricated disks in a removable 3348 data module. Keeping heads with their surfaces reduced interchange alignment requirements; Winchester did not originally require fixed media.

    Why it matters Keeping heads with their disk surfaces reduced the alignment demands of interchangeable media.

    History & engineering

    Historical context

    Earlier interchangeable disk packs required a drive's heads to align with recording surfaces that could arrive from another drive. The IBM 3340 approach reduced that interchange problem by treating the heads and their associated disks as a paired assembly.

    How it works

    The removable 3348 data module kept low-mass heads with lubricated recording surfaces. The engineering change was therefore more specific than putting a conventional disk pack inside a box: the head-surface relationship travelled with the module.

    Limits & interpretation

    Winchester is often used as shorthand for later fixed-disk designs, but that shorthand hides the original removable module. The important legacy is the integrated head-and-media approach, not a claim that every later drive used the same packaging or mechanics.

    Sources (1) Documented milestone
  3. 07

    Thin-film recording heads enter IBM drives

    IBM's engineering retrospective dates the first introduction of its thin-film inductive recording heads to 1979, after development beginning in the 1960s.

    Why it matters Fabrication techniques became central to improving the tiny structures that write and read magnetic data.

    History & engineering

    Historical context

    Magnetic recording improvements required advances in the structures that generate and sense a magnetic field, not only changes to the disk coating. IBM's engineering retrospective traces thin-film head development from work in the 1960s to its 1979 introduction.

    How it works

    Thin-film fabrication made the recording head a manufactured microscopic structure. This entry concerns an inductive head, in which the coil and magnetic structure perform recording and sensing functions; it is not the later giant-magnetoresistive sensor milestone.

    Limits & interpretation

    The distinction matters when comparing generations: a fabrication method and a sensing effect are different innovations. Thin-film development formed part of the route toward smaller recording structures, but this introduction date does not establish a universal density or performance gain for every drive.

    Sources (1) Documented milestone

1980–1999

The personal-computing era

Optical discs, flash cells and shared interfaces develop alongside disk arrays.

  1. 08

    Compact Disc reaches listeners

    Sony launched its CDP-101 player in Japan on 1 October 1982 alongside recorded music. This marks an audio product launch, not the later standardization of CD-ROM file systems.

    Why it matters Optical discs established a digital distribution medium that would also carry computer data.

    History & engineering

    Historical context

    The launch brought together a player and a catalogue of recorded music, following joint format development by Sony and Philips. The relevant milestone is a consumer audio system becoming available in Japan, not merely a laboratory demonstration of digital sound.

    How it works

    The player recovered digitally encoded music optically from the disc. A usable format required agreement across disc manufacture, playback electronics and recorded content; neither a laser nor a digital recording alone was sufficient to make an interoperable product.

    Limits & interpretation

    Audio CD should not be confused with a computer filesystem. The later use of similar optical media for files required additional organization and interchange rules. Keeping these layers separate explains why the CD launch and ECMA-119 occupy different places on this timeline.

    Sources (1) Documented milestone
  2. 09

    Flash memory is announced

    KIOXIA's retrospective dates Toshiba's announcement of flash memory to 1984, describing an electrically erasable non-volatile memory that could erase data in bulk.

    Why it matters Bulk erase was part of the path from non-volatile memory cells to dense storage.

    History & engineering

    Historical context

    Non-volatile semiconductor memory retained information without continuous power, but practical storage also needed an economical way to replace that information. Toshiba's flash work made bulk electrical erasure a defining part of the design.

    How it works

    Programming and erasing change the charge stored in a memory cell. That charge affects how the transistor conducts when read. The word flash refers to erasing groups of cells, not to a claim that every operation is instantaneous.

    Limits & interpretation

    The 1984 announcement predates the separate NAND milestone. It is not evidence that modern flash SSDs were already available. Cell operation, array organization and a complete drive controller are distinct parts of the subsequent storage story.

    Sources (2) Manufacturer account
  3. 10

    IBM 3480 moves tape into cartridges

    IBM introduced its cartridge-based 3480 tape family in 1984 as it moved away from reel-to-reel products.

    Why it matters Cartridges changed media handling without removing tape's sequential-access character.

    History & engineering

    Historical context

    Moving from open reels to the 3480 cartridge family changed how operators handled and exchanged computer tape. It was a packaging and system transition within magnetic recording, rather than a replacement of tape by a different storage medium.

    How it works

    The cartridge contained the tape supply as a removable unit while the drive remained responsible for transporting the medium past the heads. The data still occupied positions along a long strip; the cartridge did not eliminate the need to move that strip.

    Limits & interpretation

    Cartridge handling and random access are different properties. The 3480 was not the invention of tape cartridges: IBM's earlier 3850 also used cartridges. Nor was it the beginning of LTO; those are separate systems with different formats and compatibility requirements.

    Sources (2) Documented milestone
  4. 11

    ECMA-119 defines CD-ROM file structure

    The first edition of ECMA-119 appeared in December 1986. It specified volume and file structures for information interchange on CD-ROM; later work led to ISO 9660.

    Why it matters A shared logical format let different systems exchange files on the same kind of optical medium.

    History & engineering

    Historical context

    Two computers could have optical drives yet still disagree about where a disc's files began or how its directories were represented. ECMA-119 addressed that logical interchange problem rather than the optical pickup or the physical recording layer.

    How it works

    The standard describes volume information, file placement and file attributes, with defined levels of interchange. Those structures give a receiving implementation a shared way to identify files instead of relying on the originating computer's private disk layout.

    Limits & interpretation

    The December 1986 first edition is the dated event. Later editions and extensions must not be assumed to have existed in that version. A common filesystem also does not guarantee that the receiving computer can interpret every application's file contents.

    Sources (1) Documented milestone
  5. 12

    SCSI standardizes a disk bus

    ANSI X3.131-1986 (SCSI) gave workstations and servers a shared command set and parallel bus that could chain multiple devices—an alternative to ST-506/ST-412 style interfaces.

    Why it matters A common command interface could connect different kinds of peripherals, not just disks.

    History & engineering

    Historical context

    A host needed a consistent way to request operations from peripherals without implementing every drive's internal mechanics. SCSI established a common interface for intelligent devices, extending beyond the role of a dedicated hard-disk connection.

    How it works

    The interface combined commands with a bus over which a host and selected peripheral exchanged requests, data and status. Separating the requested operation from the device's internal implementation made different peripheral types possible within the same interface family.

    Limits & interpretation

    The 1986 date belongs to ANSI X3.131, not every later technology carrying the SCSI name. Later transports and revisions should not be retroactively assigned to this parallel-bus standard. The interface standard also did not define one universal drive speed.

    Sources (2) Documented milestone
  6. 13

    Toshiba announces NAND flash

    KIOXIA's technology history dates Toshiba's invention and announcement of NAND flash to 1987. The announcement is distinct from subsequent commercialization.

    Why it matters NAND became a foundation for dense non-volatile storage; an invention date is not a product launch.

    History & engineering

    Historical context

    NAND changed how non-volatile memory cells were organized for dense storage. KIOXIA separates Toshiba's 1987 announcement from later commercial products, including its 1991 NAND commercialization milestone.

    How it works

    Cells are connected in series in a NAND string. Stored charge changes a cell's threshold voltage, and the read operation distinguishes states through their conduction behavior. The structure makes the array different from a set of independently wired storage elements.

    Limits & interpretation

    A NAND chip is not a complete SSD. The later drive combines memory with a host interface and control functions. The invention date therefore establishes an array-technology milestone, not modern SSD performance, endurance or widespread replacement of hard disks.

    Sources (2) Manufacturer account
  7. 14

    RAID organizes disks into an array

    Patterson, Gibson and Katz presented the RAID case at SIGMOD in 1988. Berkeley's accompanying research examined disk-array organizations, performance and reliability.

    Why it matters Redundancy and parallel I/O could be designed at the array level, not only inside a single drive.

    History & engineering

    Historical context

    The Berkeley researchers examined whether an array of smaller disks could deliver useful capacity and I/O capability instead of relying on a single large drive. Their argument treated organization, performance and reliability as linked system-design questions.

    How it works

    Data can be spread over multiple drives so that operations use more than one device. Redundant information, placed according to the array organization, allows selected drive failures to be tolerated. Different layouts incur different capacity and update costs.

    Limits & interpretation

    Parallelism and redundancy are not interchangeable benefits. A reliability model depends on assumptions about failures and recovery, while performance depends on the access pattern. The paper is a research milestone, not evidence that every RAID level accelerates every workload or protects against every cause of data loss.

    Sources (2) Documented milestone
  8. 15

    ATA's drive interface is documented

    The 1994 revision 4c working draft documents AT Attachment: an integrated bus interface between a host and a disk drive. The cited document is a draft, not evidence that IDE first appeared in 1994.

    Why it matters The host and an intelligent drive could share a defined register and command interface.

    History & engineering

    Historical context

    The draft describes an interface between the host and an intelligent disk drive. This is a different boundary from an interface that leaves low-level drive control on a separate host controller.

    How it works

    Registers and commands provide a shared language for selecting operations, transferring data and reporting device status. Integrating more control with the drive lets the host operate through that interface rather than reproduce the drive's internal recording decisions.

    Limits & interpretation

    The surviving revision 4c document explicitly identifies itself as a working draft. Its 1994 date cannot be used as the first appearance of IDE hardware, and its documented command set should not be confused with capabilities added by later ATA revisions.

    Sources (1) Documented milestone
  9. 16

    DVD's competing proposals converge

    On 8 December 1995, the format's supporting companies announced agreement on final DVD details, following a September agreement. The specification covered movie players and computer ROM applications.

    Why it matters A common format avoided carrying two incompatible proposals into the same generation of optical media.

    History & engineering

    Historical context

    The SD and MMCD groups had promoted competing high-density optical-disc proposals. Their December statement followed a September agreement and settled details needed by both the movie and computer industries.

    How it works

    The published appendix specified a 120 mm disc, 4.7 GB per single side and a red-laser optical system. The agreement also addressed coding and error correction, showing that interchange required more than deciding the physical disc diameter.

    Limits & interpretation

    This was a format agreement, not a statement that players were already in every market. Movie and computer-ROM specifications were finalized, while music applications still required discussion. Shared physical media did not make all application formats identical.

    Sources (1) Documented milestone
  10. 17

    GMR read heads reach desktop-drive announcements

    IBM's 1997 annual report records a November announcement of a 16.8 GB desktop disk drive incorporating giant magnetoresistive heads.

    Why it matters More sensitive read heads helped detect the weaker signals associated with smaller recorded bits.

    History & engineering

    Historical context

    As recorded features became smaller, reading their magnetic signal became a distinct engineering challenge. IBM's annual report places a 16.8 GB desktop-drive announcement on 10 November 1997, connecting the sensor technology to a particular product milestone.

    How it works

    A spin-valve read sensor uses a change in electrical resistance associated with the magnetic configuration of thin layers. It senses the recorded field; this readout mechanism should not be confused with the inductive element that writes the data.

    Limits & interpretation

    The capacity and announcement date describe IBM's product, not an independent comparison of every drive available at the time. The broader lesson is that improving the read sensor can enable denser recording without making head positioning or the write process disappear.

    Sources (2) Manufacturer account

2000–2009

Flash and networked storage

Serial links, web services and vertical memory structures expand the design space.

  1. 18

    LTO-1 begins an interoperable tape lineage

    The LTO Program dates first-generation availability to 2000, with 100 GB of native capacity per LTO-1 cartridge. Native capacity excludes assumed compression.

    Why it matters An interchange format created a multi-generation path for tape backup and archival systems.

    History & engineering

    Historical context

    LTO established a tape-format lineage intended for interchange across participating suppliers. Its first generation provided a shared starting point for drives and cartridges, rather than a single vendor's isolated media format.

    How it works

    The cartridge stores magnetic tape, with the drive moving the tape to access recorded data. The LTO Program lists 100 GB of native capacity for LTO-1. Native capacity describes storage without assuming that a user's data will compress.

    Limits & interpretation

    Later LTO features should not be attributed to the first generation: the Program describes WORM, encryption and file-oriented capabilities as subsequent developments. A shared family name also does not establish unlimited compatibility between generations; preservation requires an appropriate reader.

    Sources (1) Manufacturer account
  2. 19

    Serial ATA 1.0 is released

    The Serial ATA Working Group released its final 1.0 specification on 29 August 2001. This is the specification milestone, not the later arrival of SATA drives in PCs.

    Why it matters A serial physical interface simplified the cables and connectors used by internal storage.

    History & engineering

    Historical context

    The Serial ATA Working Group sought a successor to parallel ATA's physical connection while easing system integration. The 2001 release described a specification and an industry development effort, with products still expected to follow.

    How it works

    The serial link reduced the number of signal connections and enabled smaller connectors and easier cable routing. The announcement specified an initial 1.5-gigabit-per-second signaling rate; that is an interface rate, not a measured file-copy speed.

    Limits & interpretation

    Changing the cable and signaling interface does not remove the storage device's own bottlenecks. The release therefore should not be read as a universal performance benchmark, nor should its 2001 publication date be replaced by the later date when SATA products became familiar.

    Sources (1) Documented milestone
  3. 20

    Blu-ray's basic specification is announced

    Nine companies announced Blu-ray's basic specification on 19 February 2002. A 405 nm blue-violet laser was central to its higher-density optical recording design.

    Why it matters A shorter wavelength helped reduce the optical spot size; this announcement preceded later consumer launches.

    History & engineering

    Historical context

    The nine-company announcement targeted recording high-definition video on an optical disc of familiar diameter. It concerned the basic specifications of a recording format, before the later consumer launch history.

    How it works

    A 405 nm laser worked with a lens numerical aperture of 0.85 and a thin optical protection layer. The announcement treats these as a combined optical design: wavelength alone is not the full explanation for a smaller recording spot.

    Limits & interpretation

    The initial specification listed several capacities, including 23.3, 25 and 27 GB. It should not be silently rewritten as a claim about every later Blu-ray format. The 2002 announcement also concerned rewritable phase-change recording, not just movie distribution on pressed discs.

    Sources (1) Documented milestone
  4. 21

    Toshiba announces perpendicular-recording HDDs

    Toshiba announced perpendicular-recording 1.8-inch HDDs in December 2004, with mass production planned for 2005. The announcement and the production schedule are separate milestones.

    Why it matters Recording magnetization perpendicular to the platter provided another route to denser magnetic storage.

    History & engineering

    Historical context

    Longitudinal recording oriented magnetization along the plane of the disk. Toshiba's announcement described a different head-and-media design aimed at increasing density in small 1.8-inch drives.

    How it works

    Perpendicular recording orients the stored magnetization out of the plane. Toshiba emphasized the combination of a suitable disk structure, a new head and their integration, rather than a software change applied to an unchanged drive.

    Limits & interpretation

    The release named 40 GB and 80 GB models and scheduled their mass production for separate quarters in the fiscal year starting April 2005. A planned production schedule is not itself a shipment measurement, and the announcement's priority claims remain the manufacturer's account.

    Sources (1) Manufacturer account
  5. 22

    Amazon S3 exposes storage through a web service

    AWS launched Amazon S3 in March 2006 with a web-services interface for storing and retrieving data. This is a service launch, not the invention of object storage.

    Why it matters An application could address remotely managed storage through an API instead of managing a local disk.

    History & engineering

    Historical context

    The service offered developers access to remotely operated storage without requiring them to procure and operate the underlying disks. The dated announcement marks AWS making that service available, not the invention of networked or object-based storage.

    How it works

    The application-facing operation became a web-service request to store or retrieve data. That interface boundary separates what the developer requests from how the provider implements the physical storage underneath.

    Limits & interpretation

    The launch announcement does not define every guarantee or feature S3 acquired later. It also is not a local block-device or filesystem specification. Historical comparisons must use the service contract from the relevant period rather than assuming today's behavior existed in March 2006.

    Sources (1) Documented milestone
  6. 23

    Toshiba announces 3D flash technology

    KIOXIA dates Toshiba's announcement of 3D flash technology to 2007. Its history distinguishes that work from the later commercialization of BiCS FLASH.

    Why it matters Vertical structures opened an alternative to relying only on shrinking planar cells.

    History & engineering

    Historical context

    Shrinking planar memory cells created physical and manufacturing difficulties. Toshiba's 2007 work explored adding cells vertically, while also addressing the cost of fabricating a stack one complete memory layer at a time.

    How it works

    BiCS uses stacked electrode and insulating layers, with holes passing through them. Charge-storage material and vertical structures are formed in those holes, allowing memory cells at multiple levels to be created through shared processing steps.

    Limits & interpretation

    The architecture announcement is distinct from a production device. KIOXIA's account dates 48-layer BiCS commercialization to 2015. Nor is a vertical cell array simply a package containing several separate chips; these are different ways to increase capacity.

    Sources (2) Manufacturer account
  7. 24

    Intel launches X18-M and X25-M SSDs

    Intel announced its X18-M and X25-M SATA solid-state drives in September 2008 for notebook and desktop computers. Flash SSDs already existed before this launch.

    Why it matters Flash storage was reaching familiar PC drive interfaces and form factors.

    History & engineering

    Historical context

    Intel's launch placed MLC NAND drives into the 1.8-inch and 2.5-inch form factors used by PCs. The announcement said 80 GB products were shipping, while 160 GB versions were expected to sample later.

    How it works

    The drive combined flash with a controller, firmware and a parallel ten-channel architecture. Intel explicitly discussed wear leveling and write amplification, illustrating that an SSD is a managed storage system rather than a passive collection of memory chips.

    Limits & interpretation

    The quoted performance figures came from Intel and its stated tests. They should not be turned into a workload-independent speedup or a fleet reliability result. This launch is one stage in client SSD adoption, not the first appearance of solid-state storage.

    Sources (1) Documented milestone

2010–2024

New paths to density and scale

NVMe, 3D NAND and assisted magnetic recording evolve in parallel.

  1. 25

    NVMe 1.0 defines a new storage interface

    NVM Express released version 1.0 on 1 March 2011, with a command set and parallel queue model for PCIe-attached non-volatile memory. Product adoption followed the specification.

    Why it matters Storage gained a command and queue model designed for non-volatile memory on PCIe.

    History & engineering

    Historical context

    Fast non-volatile memory exposed the cost of interfaces designed around older storage assumptions. NVMe defined a host-controller interface for PCIe storage, addressing the command path as well as the underlying link.

    How it works

    Separate submission and completion queues organize requests and their results. Multiple queues support parallel work from modern systems, instead of forcing all activity through one shared command path. The protocol is distinct from the physical form factor of a drive.

    Limits & interpretation

    More queue capacity does not make a lightly loaded request faster by itself. Device behavior, software and workload still determine observed performance. The March 2011 specification release is therefore an interface milestone, not proof that all systems immediately adopted NVMe.

    Sources (1) Documented milestone
  2. 26

    Seagate reports SMR drive shipments

    In September 2013, Seagate reported that it had shipped more than one million SMR drives. The shipment count is the manufacturer's statement, not an independent market measurement.

    Why it matters Overlapping tracks increased density while making updates to existing data more complicated.

    History & engineering

    Historical context

    Seagate's shipment statement marked a commercial step for an approach that sought higher capacity by placing magnetic tracks closer together. The reported million-drive count is useful historical evidence of the company's claim, not an audited industry total.

    How it works

    An SMR write overlaps part of a neighboring track while leaving a narrower readable region. Rewriting older data can therefore affect adjacent recorded data. Bands and their management provide a way to handle that dependency.

    Limits & interpretation

    Drive-managed and host-managed SMR expose different responsibilities to software. Capacity gains cannot be converted into a blanket write-performance claim: behavior depends on how updates, free space and sequential writing are handled. A shipment milestone alone does not answer those workload questions.

    Sources (2) Manufacturer account
  3. 27

    Samsung reports V-NAND mass production

    Samsung announced mass production of 24-layer V-NAND in August 2013. This is a manufacturer's production report, distinct from earlier 3D-memory research announcements.

    Why it matters Stacking cells added a vertical dimension to NAND density scaling.

    History & engineering

    Historical context

    Samsung's announcement addressed the difficulties of continuing to shrink planar NAND. The company reported mass production of a 128-gigabit device with 24 vertically stacked cell layers, moving the discussion from research structures to manufacturing.

    How it works

    The design used charge-trap storage and vertical interconnections. Increasing the number of cell layers provides another way to increase density; layer count and the number of bits represented by each cell are different dimensions of a memory design.

    Limits & interpretation

    The reported reliability and write-speed improvements were Samsung's comparisons against specified planar technology. They are not independent fleet results or guarantees for all later 3D NAND. A production report also does not mean every SSD immediately incorporated the new memory.

    Sources (1) Manufacturer account
  4. 28

    NVMe over Fabrics extends the interface beyond PCIe

    NVM Express announced its first NVMe over Fabrics specification in June 2016. The layered architecture allowed NVMe commands to travel across network fabrics.

    Why it matters Storage and compute could be separated while retaining the NVMe command model.

    History & engineering

    Historical context

    The original local PCIe interface did not by itself provide a way for many hosts to reach storage across a data center. NVMe over Fabrics extended the command model beyond that local attachment boundary.

    How it works

    The 2016 release described a layered architecture and Linux reference implementations for RDMA and Fibre Channel transports. A network transport carries the NVMe operations, allowing compute and storage placement to be separated.

    Limits & interpretation

    Preserving the command model does not remove network delay, congestion or failures. Transport versions and implementations matter, so later NVMe/TCP support should not be silently folded into this initial release. The milestone is a specification, not a universal remote-storage latency guarantee.

    Sources (1) Documented milestone
  5. 29

    Zoned Namespaces expose placement constraints

    NVM Express ratified the Zoned Namespace command set in June 2020. A namespace can be divided into zones with explicit write and management rules.

    Why it matters Host software gains placement responsibilities; a zoned interface is not a drop-in performance guarantee.

    History & engineering

    Historical context

    Conventional block interfaces conceal much of the device's placement and reclamation work. ZNS exposed a different contract in which host software participates in organizing writes into independently managed zones.

    How it works

    A zone has a write position and usable capacity. A write to a full zone fails; the host must select another suitable zone rather than assume automatic spillover. Zone management is therefore part of the software's normal storage path.

    Limits & interpretation

    The interface can support better cooperation between applications and flash, but it requires compatible software. Zone size and usable capacity are not necessarily equal, and advertised benefits depend on workload and implementation. Ratification does not make ZNS a transparent upgrade for every filesystem.

    Sources (1) Documented milestone
  6. 30

    Mozaic 3+ brings HAMR to a new product generation

    On 17 January 2024, Seagate announced Mozaic 3+ and a quarterly 30 TB-plus hyperscale volume ramp, with some customer qualification still pending. This is a vendor product announcement, not universal availability.

    Why it matters Heat-assisted writes address smaller, magnetically stable bits. Seagate had already reported 20 TB HAMR shipments in calendar 2020; the 2024 launch is a later product milestone.

    History & engineering

    Historical context

    Smaller magnetic grains can increase density but also make stable retention harder. Mozaic 3+ combined high-coercivity iron-platinum media with an assisted writing process, tackling both the ability to retain a bit and the ability to change it.

    How it works

    A laser and near-field transducer deliver localized heat at the write location, temporarily making the medium easier to switch. This is an aid to magnetic writing, not optical data storage like a CD or Blu-ray disc.

    Limits & interpretation

    Seagate's January 2024 statement described a hyperscale ramp and customer qualification, not unrestricted availability or an independent lifetime study. Its earlier report of 2020 HAMR shipments also means that Mozaic 3+ should not be labeled the first-ever HAMR shipment.

    Sources (3) Manufacturer account

Scope & evidence

One date does not mean one kind of milestone

A research announcement, working draft, published standard and product launch are different events. Each entry names the kind of milestone being dated. Manufacturer accounts are identified explicitly; an announcement does not establish independent performance, market adoption or historical priority.

This is a curated chronology, not an exhaustive inventory. Entries are spaced for reading, not proportionally to elapsed time. Each event has a stable link, source references and, where available, related engineering coverage. Source access dates are not event dates.

Definitions and machine-readable knowledge at hesela.dev