MEMS-BASEDINTEGRATED-CIRCUITMASS-STORAGE SYSTEMS: MEMS-BASED INTEGRATED-CIRCUIT MASS-STORAGE SYSTEMS L. R. Carley, G. R. Ganger, D. F. Nagle Carnegie-Mellon University


Paper highlights: Paper highlights Discusses a new secondary storage technology that could revolutionize computer architecture Faster than hard drives Lower entry cost Lower weight and volume Lower power consumption Paper emphasis is on physical description of device


DISK DRIVE LIMITATIONS: DISK DRIVE LIMITATIONS Disk drive capacities double every year Better than the 60% per year growth rate of semiconductor memories Two major limitations of disk drives are Access times decreases have been minimal Minimum entry cost remains too high for many applications


Stating the problem: Stating the problem We need a type of new mass storage that can break both barriers of Access times Minimum entry cost New mass storage should also be significantly cheaper than non-volatile RAM $100 now buys 1 GB of flash memory


MEMS: MEMS Microelectromechanical systems (MEMS) use Same parallel wafer-fabrication process as semiconductor memories Keeps the prices low Same mechanical positioning of R/W heads as disk drives Data can be stored using higher density thin-film technology


Main advantages of MEMS (I): Main advantages of MEMS (I) Potential for dramatic decreases in Entry cost Access time Volume Mass Power dissipation Failure rate Shock sensitivity


Main advantages of MEMS (II): Main advantages of MEMS (II) Integrate storage with computation Complete systems-on-a-chip integrating Processing unit RAM Non-volatile storage Many many new portable applications


THE CMU MEMS PROTOTYPE: THE CMU MEMS PROTOTYPE Like a disk drive, it has recording heads a moving magnetic recording medium Major departures from disk drive architecture are MEMS recording heads—probe tips—are fabricated in a parallel wafer-level manufacturing process Media surface does not rotate


How the media surface moves: How the media surface moves Media surfaces that rotate require ball bearings Very small ball bearings have “striction” problems that prevent accurate positioning Elements would move by sticking and slipping Best solution is to have media sled moving in X-Y directions Sled moves in Y-direction for data access Sled is suspended by springs


Conceptual view: Conceptual view Sled with magnetic coating on bottom Fixed part with tip array Sled suspension is omitted from drawing


The media sled: The media sled Size is 8mm x 8mm x 500 mm Held over the probe tip array by a network of springs Motion applied through electrostatic actuators Motion limited to 10% or less of suspension/actuator length Each probe tip can only sweep 1% of the media sled


The probe tip array: The probe tip array Includes a large number of probe tips for Being able to access whole media sled (in combination with X-Y motions of sled) Improving data throughput Increasing system reliability


Probe tip positioning (I): Probe tip positioning (I) Most MEMS include some form of tip height control because Media surface is not perfectly flat Probe tip heights can vary CMU prototype places each probe tip on a separate cantilever Cantilever is electrostatically actuated to a fixed distance from the media surface


Probe tip positioning (II): Probe tip positioning (II) IBM Millipede Uses a 32 x 32 array of probe tips Each tip is placed at the end of a flexible cantilever Cantilever bends when tip touches surface HP design places media surface and probe tips sufficiently apart No need to control probe tips


Probe tip positioning (III): Probe tip positioning (III) CMU solution is most complex of three Must control individual heights of 6,400 probe tips Required by recording technology


Probe tip fabrication: Probe tip fabrication Major challenge is fabricating read/write probe tips in a way that is compatible with the underlying CMOS circuitry This includes thermal compatibility geometrical compatibility chemical compatibility ...


Media positioning: Media positioning System’s current target is to have each probe tip in the middle of a 100 mm square Media actuator must be able to move at least ±50 mm in each direction Can be achieved with an actuation voltage of 120V Well above CMOS rated voltage


Storing, reading and writing bits: Storing, reading and writing bits CMU prototype uses same magnetic recording technology as current disk drives Minimum mark size is around 80mm x 80mm Other solutions include Melting pits in a polymer (IBM Millipede): Raises tip wear issues Phase change media (HP prototype) Same technology as CD-ROM


PROTOTYPE PERFORMANCE (I): PROTOTYPE PERFORMANCE (I) All data were obtained through simulation Average service time around 0.52 ms Disk drive service time is 10.1 ms Key factor for service time is X-seek time I/O bandwidth depends on number of simultaneously active tips per-tip data rate


PROTOTYPE PERFORMANCE (II): PROTOTYPE PERFORMANCE (II) Sustainable data rate is not a linear function of access data rate Track switching time now depends on access velocity: Faster sled means higher turn around time Maximum sustainable data rate of single tip varies from 1.4 to 1.8 Mb/s Reached for peak data rate of 2 to 3 MB/s


Application performance: Application performance PostMark benchmark: Models file activity in Internet servers Prototype is 3.4 times faster than current drives Much faster metadata updates TPC-D benchmark: Models transaction processing Prototype is 3.9 times faster despite extensive caching in competing disk drive


POTENTIAL APPLICATIONS: POTENTIAL APPLICATIONS Lighter and less shock sensitive than disk drives Great for notebook PC’s, PDA’s and video camcorders Lower cost than disk drives in 1 to 10 GB range Will open many new applications High areal densities Great for storing huge amounts of data Can combine computing and storage on a single chip


MY OVERALL OPINION: MY OVERALL OPINION Technology has a bright future if and when production kinks get solved We should remain somewhat skeptical Not the first “gap-filling” technology to be tried Bubble memories were “hot” in the 70’s Lower RAM prices killed them in the early 80’s Watch prices of non-volatile RAM