Key contributions to the data recording technology of modern hard disk drives

Magnetic Data Recording for Hard Disks

Harddisks are indispensable components of any computer system from the smallest mobile PC all the way up to the biggest server system. They are used as secondary storage to safely hold system and user data and programs even at times when the computer is switched off. Hard disks are also the storage device of choice in an increasing variety of consumer products that require massive amounts of cheap non-volatile storage, such as digital video recorders, MP3 players, mobile phones, or digital cameras. Storage density and data rates of hard disks have grown at an astounding pace during the past decades; at the same time, cost per information unit stored on the disks magnetic media has decreased dramatically. This was made possible bye continuous technical advances of all components of the system: recording heads, storage media, disk drive mechanics and electronics. As part of the drive electronics, the recording channel detects the analog signal coming from the read head of the drive and converts it into a bit sequence. At the high storage densities and data rates of modern disk drives, reliable signal detection representes a formidable task because the signal usually is extremely weak, distorted, and buried in noise coming from the magnetic media, the electronics, and neighboring data. Traditionally, recording channels in hard disks used a peak detection scheme to recover the correct bits from the read signals. This straightforward technique served the disk drive industry well in the first three decades of its existence.

It was Hisashi Kobayashi who in his 1970/71 papers pointed out that a digital magnetic recording channel can be viewed as theoretically equivalent to a baseband data transmission system with “partial-response” signaling and that the Viterbi algorithm can be used to obtain maximum likelihood sequence detection of the symbol sequence that has been coded by partial-response channel signaling. This dynamic programming alogrithm, named after its developer Andrew Viterbi (Winner of the 1994 Eduard Rhein Basic Research Award), had originally been conceived for decoding convolutional codes. Kobayashi s analysis indicated that a detector based on the above principles would have significant advantages over the bit-by-bit decision technique employed in the conventional peak detection technique.

As it is often the case with groundbreaking insights, Kobayashi s idea was ahead of its time: Initially, no one in industry or academia pursued in further until, in the second half of the 1970s, François Dolivo systematically investigated signal processing alternatives for disk drives including an approach in which the read signal takes the form of a partial-response class IV (PRIV) signal. The investigations revealed significant potential gains in recording density and provided the motivation to verify and demonstrate the technique in a prototype system. In subsequent years, Dolivo and his team at IBM s Zurich Research Laboratory, together with the disk drive developement organization in Rochester, MN, devised and developed all necessary elements of a new recording technology which they called “Partial Response Maximum Likelihood Sequence Detection” (PRML). In 1990, their work culminated in the announcement of the industry s first hard disk drive product using the new channel. PRML, which rapidly became the de-facto industry standard, boosted storage density and data rates enabling for a number of years compound annual growth rates of 60% and 40%, respectively.

In the second half of the 1990 s, the continued increase of areal density an data rates dictated a departure from the existing read channel technology. In response to this need, Evangelos Eleftheriou came up with the concept of “Noise-Predictive Maximum Likelihood” (NPML) detection for magnetic recording. The starting point of this method was the observation that the noise which accompanies the useful signal coming from the disk s read head is not completely random, as assumed until then. To account for this effect, NPML introduced a noise prediction/ whitening process into the computation algorithm of the detector. NPML combined with additional sophisticated signal processing and coding techniques enabled further substantial increases in linear recording density. The new architecture was introduced into IBM disk drive products in 2000. Today, NPML and variations thereof have become the accepted recording technique in the hard disk industry.

By presenting its 2005 Technology Award to Hisashi Kobayashi, François Dolivo, and Evangelos Eleftheriou, the EDUARD RHEIN FOUNDATION honors three pioneers of magnetic recording whose contributions to the field were key to the unique progress of hard disk drives during the past decades.

Dr. Sönke Mehrgardt,
SODANA CONSULTING, Deisenhofen