The future of tape for data storage: the 1-TB cartridge and beyond.

Multiple factors are driving increased demands on storage--particularly tape storage. Advances in Business Continuance planning and compliance with more stringent legal restrictions are driving data growth to higher and higher levels. Management capabilities for tape storage are also reaching new levels with advances in Virtual Tape solutions. With both the management facilities and the growth requirements expanding, the question is "Can tape technology meet the challenge?" The challenges being to both meet increasingly stringent requirements, as well as outpace competing technologies such as disk. A look at the technology drivers and roadmaps shed light on the question.

[FIGURE 1 OMITTED]

[FIGURE 2 OMITTED]

Storing and retrieving data on magnetic tape is driven by (a) capacity (GBytes/cartridge) primarily because of the cost of storage ($/GByte), (b) data rate (MBytes/second) as people don't want to wait forever and (c) reliability (the data has to be there!). The capacity of a tape cartridge or a disk is simply the areal density of the data multiplied by the area of the media used but is often preferably computed in tape by using the relation

C = NbL[epsilon]/8

in bytes, where N is the number of tracks across the tape, b is the linear recording density in bits per inch, L is the length of the tape (in inches) and [epsilon] is a formatting/ECC (error correction code) overhead efficiency factor (typically about 0.7). The 8 assumes 8 bit bytes. The date rate is given by

D = nbV[epsilon]/8

in bytes/second, where n is the number of parallel channels used and V is the speed of the tape (inches/second). These two relations capture the main parameters in increasing capacities to terabyte levels and data rates to 100s of Mbytes/sec. The linear density (b) appears in both calculations and thus is a strong contributor to the technology. The data reliability is assured as the data is encoded with powerful ECC systems that spread the data both down the length of tape and across the width with the user data shared among the many multiple parallel channels. This provides protection against any media or other defects on any given track or channel or even combinations thereof. This gives orders of magnitude improvement in data error rates compared to single channel systems such as magnetic disk drives.

Table 1 shows scenarios for 0.5, 1, 5 and 10-terabyte capacities for various data rates for a normal IBM3480/STK9940/LTO/DLT half-inch wide tape cartridge form factor. Some tradeoffs between the parameters given in equations 1 and 2 have been included for illustrative purposes and one can easily see where a different set of tradeoffs could yield the same result, depending on which aspect of the tape system you wish to stress more.

Some of the dimensions and time scales may seem aggressive but are well within the capability of technologies that either exist or are under development. Fundamental to recording digital data on magnetic tape is the analog magnetic recording that takes place between the head and the media. These two magnetic components in combination can make or break a reliable data recording system.

Figures 1 illustrates magnetic recording on tape and its digital interpretation. The digital interpretation is that a transition between regions on the tape magnetized in one direction to the opposite direction is interpreted as a logical '1' and the absence of the transition a '0' when referenced to a data clock. This interpretation depends on the logic used by the detection system and coding design. For instance a PRML (Partial Response Maximum Likelihood) channel interprets the recorded transitions in a different way by partial amplitude sampling in order to increase the bit density using somewhat lower magnetic transition densities than in straight peak detect channels as illustrated. Such channels increase the logical bit density up to twice that of the recorded magnetic transition density and are already in use in tape systems today.

[FIGURE 3 OMITTED]

Fundamentally, an increase in linear recording density requires the transitions to be closer and closer together on the media together with the ability to resolve them. Table 1 indicates the length of a logical bit (bit cell in nanometers (nm)) for the various scenarios given for reference (~50-100nm). Tape media to date has had the magnetic coating somewhat thick (0.5[micro]m or more) compared to these dimensions combined with moderate magnetic coercivity which yields broad written transitions due to the generation of transitions curving into the depth of the magnetic coating and the demagnetizing effect of sizeable opposing magnetic regions. Figure 2 shows how linear density has indeed gated tape products in the past according to media coercivity together with a projection for future systems based on published roadmaps (the data is taken from existing IBM. STK, Quantum DLT and LTO tape products and roadmaps).

From this piece of fundamental physics, the tape industry knows that a coercivity increase and a reduction in the magnetic thickness of the media is the primary direction to pursue as already demonstrated by magnetic disk technology. Recently, this has been achieved in particulate media by using a dual coating process. Here the magnetic portion of the tape coating is spread thinly over a simultaneously coated non-magnetic underlayer. This effectively provides a thick physical coating for smoothing purposes coupled with a reduced magnetic thickness. This has enabled coatings to be produced as low as 100nm and progress is being made to reduce this further (Fujifilm NanoCubic tape and Maxell NanoCap). There is also ongoing research into advanced thin film tape media, which will continue the density growth as was seen in magnetic disk. (Tape is indeed fortunate that magnetic disk has already demonstrated solutions to high areal density magnetic recording while being able to leverage that for higher volumetric densities.)

The other avenue that tape has room to grow is in track density. Significant advances in track following servo capability to track narrow recorded tracks are enabling the track densities listed in Table 1. Head structures and architectures for multi-channel operation for high data rates are not approaching any fundamental limits at this time. Head technology appears to have enough precedents and product introductions (again as seen in disk magnetic recording) that tape head offerings should be able to readily respond to new media types as they are developed. A classic example would be the shift to all thin film write heads and thin film shielded read heads as well as merged pole/shared shield structures commonly used in disk and now being seen in tape applications. Using a write wide read narrow scenario, as linear tape currently does, and invoking advanced channels and ECC systems the data reliability in a tape device can meet the demands of today's and tomorrow's data centers.

The areal densities in the cases shown in Table 1 approach 5Gb/sq. in. which is 10 or 20 times lower than today's magnetic disk drives and this moderate areal density leads to volumetric densities of over a Tb/cubic inch. Figure 3 shows the volumetric density growth that tape has been seeing and is predicted by industry experts. The INSIC (INformation Storage Industry Consortium) magnetic tape roadmap of 2002 shows a 60% compound annual growth rate as indicated.

Conclusions

It is clear that ongoing advances in the medium, heads and channels are providing an ongoing volumetric density growth in tape storage leading to ongoing improvement in the cost of storage ($/GByte) for tape. This is in contrast to the story in magnetic disk where the areal density growth has slowed significantly with the associated slow down in the improvement in its cost of storage metric. The scenario of disk becoming cheaper than tape is not playing out as projected a few years ago. The disk-tape price convergence has stopped and is now beginning to diverge with tapes cost advantage growing. The fundamental physics problems that has slowed the areal density growth in disk is not on the foreseeable horizon for tape technology as volumetric density of the media in the cartridge is the driving metric not solely areal density on the media surface. Given the rather moderate areal densities currently seen in tape systems and optimism with regard to the development of tapes with thinner magnetic coatings, data storage systems using tape are poised to make some rapid advances in capacity and data rate.

Table 1. Terabyte Operating Points

Capacity (TB)                          0.5      0.5      1        1
Data Rate (MB/sec)                    60      120      110      220

No. of Pll Data Channels, n           16       32       16       32
No. of Data Tracks. N                768      768     1344     1344
Trk. Pitch (m)                        14.0     14.0      8.0      8.0
Channel Pitch, [c.sub.p] (m)         109       55      109       55
Rd. Track Width (m)                    7.0      7.0      4.0      4.0
Tape Speed, V (m/s)                    4.8      4.8      8.0      8.0
Bit Density (kbpi)                   224      224      248      248
Track Density (tpi)                 1812     1812     3172     3172
Areal Density (Gb/i[n.sup.2])          0.41     0.41     0.79     0.79
Volumetric Density (Tb/i[n.sup.3])     1.35     1.35     2.83     2.83
Bit Cell (nm)                        114      114      103      103
Bit Cell (ns)                         23.7     23.7     12.9     12.9
Tape Length (m)                      865      865      865      865

Capacity (TB)                          5        5       10       10
Data Rate (MB/sec)                   175      350      280      560

No. of Pll Data Channels, n           16       32       16       32
No. of Data Tracks. N               4140     4140     4140     4140
Trk. Pitch (m)                         2.6      2.6      2.6      2.6
Channel Pitch, [c.sub.p] (m)         109       55      109       55
Rd. Track Width (m)                    1.3      1.3      1.3      1.3
Tape Speed, V (m/s)                    9.1      9.1     10.0     10.0
Bit Density (kbpi)                   343      343      500      500
Track Density (tpi)                 9771     9771     9771     9771
Areal Density (Gb/i[n.sup.2])          3.35     3.35     4.89     4.89
Volumetric Density (Tb/i[n.sup.3])    13.15    13.15    26.39    26.39
Bit Cell (nm)                         74       74       51       51
Bit Cell (ns)                          8.1      8.1      5.1      5.1
Tape Length (m)                     1000     1000     1400     1400

Table 3: Comparing BC, DR and ILM

By Dr. Richard H. Dee

Dr. Richard H. Dee is a senior fellow working on Advanced Digital Tape Recording Technology at StorageTek (Louisville, CO)

www.storagetek.com