Analyses / Magnetic recording
Magnetic grain models: what the tiles actually represent
/ Source review and corpus correction
A rectangular tile in a magnetic-recording model represents a grain, not a new way to manufacture a disk. To interpret a result, identify how the medium was generated, how writes overwrite grains, and what information the decoder was allowed to use.
On this page6 sections
An unsupported definition, corrected
Hesela's tiled-media record previously described a patterning technique that groups grains into tiles. That claim lacked supporting evidence. The record now carries an explicit correction and unknown status. Its identifier is retained; it has not silently become a different concept.
The phrase is an unreliable search key. An English patent translation uses tiled in a title while its description names shingled magnetic recording. Another patent, titled Tilted magnetic recording media, uses tiled in claim 6 while describing tilted magnetic easy axes. Neither establishes the withdrawn grain-grouping claim. [Patent translation] [Tilted media]
The separately indexed discrete-grains model is a documented research concept. It is not an asserted synonym for that ambiguous legacy label.
Geometry, fabrication and track layout
| Concept | What the shape means | Evidence |
|---|---|---|
| Rectangular-grain model | A mathematical representation of grains and overwrite behavior. | Model, section II |
| Bit-patterned media | Fabricated magnetic islands, with one bit per island. | Fabrication review, section I |
| Shingled recording | Overlapping written tracks; not rectangular grain fabrication. | Patent description |
The fabrication review distinguishes islands defined by lithography from conventional granular media, where a recorded bit spans an ensemble of grains. Matching the appearance of a diagram is not enough to identify a recording technology. [Review, section I]
The grain mix does not specify the random medium
Das and Kashyap show why tile proportions alone leave the distribution over complete tilings unspecified. Their Metropolis-Hastings approach makes the sampling target explicit. But the fixed-count local-move chain can have isolated states: a stationary distribution is not proof that every admissible arrangement is reachable. Their broader mixing-time discussion also warns that consecutive samples are dependent. [Sampling paper, sections I, IV-VI]
For reproduction, request the generator, boundary rules, initialization and sample-spacing policy alongside the grain proportions. A seed without the sampling procedure does not define the experiment. These are checks on the model's evidence, not a claim that real disks have rectangular grains.
Read the simulation controls, not just the gain
Carosino and colleagues' model gives a grain the polarity of the last bit written in raster order. Several bit positions inside a grain therefore need not preserve independent values. Their evaluation targets a bit error rate of 10-5 or lower. [Detector paper, sections II and IV]
Two details constrain reproduction. When each two-cell orientation's probability exceeds about 0.36 (P2 = P3), the generator assembles the medium from 16 x 16 sub-images because the greedy large-image procedure struggles. Decoder iterations stop by comparison with the known transmitted block, information unavailable to a deployed receiver. The authors acknowledge the latter limitation. [Sections IV.A-B]
These choices do not by themselves invalidate the study. They do mean that a reproduction must preserve or explicitly vary both the sampling procedure and the stopping oracle. A simulated coding gain is not a measured increase in shipping drive capacity.
Use the result at the right level
This review corrects terminology and identifies consequential experimental controls. It does not reproduce the papers' algorithms or estimate HDD performance. No paper figure, dataset or implementation is republished.
For a separate physical question, continue with magnetic retention time and probability. Geometry, retention, track layout and detection each require their own assumptions.
Primary sources
- Das and Kashyap. MCMC Methods for Drawing Random Samples from the Discrete-Grains Model of a Magnetic Medium; sections I, III-VI.
- Carosino et al. Iterative Detection and Decoding for the Four-Rectangular-Grain TDMR Model. arXiv:1309.7518v1 (2013), sections II-IV.
- Albrecht et al. Bit Patterned Magnetic Recording: Theory, Media Fabrication, and Recording Performance. arXiv:1503.06664v1 (2015), section I.
- US20070009766A1: Tilted magnetic recording media, abstract and claims 1-7.
Sources checked 20 October 2026. Model and sampling papers were inspected beyond their abstracts; the fabrication citation is limited to its introductory mechanism comparison.