Cylinder Drive Concepts

This web page describes cylinder drive concepts where a cylinder is spun in a sealed vacuum containment enclosure using passive magnetic bearings at very high RPM.

Tiny memory components discussed here would be used with a powerful error-correcting system so that they do not have to be very reliable to be practical.  The error-correction system would tolerate many component failures.

The following Table compares a disk drive with several cylinder drives.

 Recording Object Disk CylinderCylinder Cylinder 
 Diameter 2.5 in0.5 in1.0 in 1.0 in 
 Square Inches of Recording Surface Area7.36 7.36 7.36 7.36 
 Length6 in 3 in3 in 
 Capacity300 GB300 GB 300 GB 300 GB 
 Average Latency3 msec 0.15 msec 0.15 msec 75 usec 
 RPM 10,000 200,000200,000 400,000 
 GB per cubic inch 46.5 88.9 64.064.0 
 Power Consumption ~7.5 W < 1W (est) < 1W (est) < 1W (est)

Cylinder Bearings

The passive magnetic bearings consist of two rings of strong permanent magnet material (such as neodymium iron boron or "Neo" for short).   A cross section view of the two rings is shown below.



The rings are magnetized as shown below so that when the inner ring is placed inside the outer ring, they will repel to form a passive magnetic bearing.



To assemble the bearing, the outer ring and enclosure would be cut into two pieces and then cemented together as illustrated below.




The two inner rings are attached to a recording cylinder as shown below.   The brown represents magnetic coating on the cylinder’s outer surface. The following drawing is close to an actual size showing the cylinder free to spin.




The cylinder probably would not have enough mass to cause a bearing crash if the drive was dropped, but, if crashing of the bearing is a problem, that problem can be solved in several different ways. One way is to combine active and passive bearings. The illustration below shows one possible configuration.




The electromagnets would be inexpensive and could easily be embedded in the enclosure. They would, most-likely, require very little current.  The active bearings would be the fail-safe backup in the event the drive was exposed to any sudden sharp shock.

Combining the active and passive bearings allows the head to recording surface distance to be controlled by moving the cylinder rather than the head.

Iron rings could be bonded to the cylinder to keep the passive and active bearings separate.

Another way to solve a bearing crashing problem is to use auxiliary bearings (probably some type of sleeves) as fail-safe bearings if the passive bearing fails.  The fail-safe bearing would only need to work maybe 10-15 times before it would be subject to failure itself.  It would only come into use if the unit was dropped.  That type of scheme is used with the Inductrack Maglev train.

Motor

The gray center ring in the drawing below is the motor magnet.





The motor magnet would be magnetized as shown below.


Coils for the motor would be attached to an insert which is attached to the end cap and slid inside the cylinder. The insert could be inserted into an accommodating slot on the other end cap for rigidity.


Air would be pumped out of the enclosure, the enclosure sealed, and the cylinder spun at possibly several hundred thousand RPMs.

Fixed Heads

In the drawing below, the red areas represent fixed recording heads.  One way to implement this idea would be to use one head per track and implement multiple heads on one IC.  Eliminating the head positioning system would eliminate a lot of cost and complexity and allow for much higher performance.  In that case, the only mechanical motion affecting access time is the rotation of the cylinder and, at 300,000 RPM, the average access time would be around 0.1 msec or 100 usec.



Moving the Head Axially

In the following drawings, the pink pieces are Neo magnets, the blue pieces are iron.

Neo magnets can be magnetized any way as long as a magnetization fixture can be built.

Consider the following Neo magnets magnetized as shown below.


 
 

When the smaller solid triangular piece is placed inside the larger piece, it forms a linear, passive magnetic bearing.  The flux lines are not always shown in subsequent drawings when is it clear how the magnets are magnetized.

The solid triangular piece has very little mass so it would take very little force to move it in a vacuum.

It should be straightforward to magnetize the triangular piece of Neo (or other similar magnetic material) so that each triangular side is either a N or S pole as illustrated below.





The triangular piece can be moved laterally by an electromagnet as shown below.



 The following drawings show how a head could be moved axially.





Head to Surface Positioning

Automatic Head to Surface Positioning

When watching a game of badminton, Burt Rutan observed that the shuttlecock always automatically positioned itself.  He used that idea to design his SpaceShip aircraft so that it automatically positions itself in flight without pilot intervention.   Air bearings similarly automatically position a head with respect to the surface, and the following idea would also automatically position the head with respect to the surface.

The following drawings show how end ring magnets and compression springs could be used to “fly” a head in a vacuum.   The springs would provide a downward force balanced by the upward force of the magnet.  When the surface rises the attraction force on the magnet increases which rises the head.

Here is an end view with a ring magnet on the end of a cylinder.  There would be two dole-pineapple-shaped permanent magnets on each end of the cylinder with the poles on the curved surfaces.  Flux from the magnets would be carried by the iron piece.  The drawings after the following drawing are views of a cylinder inside an enclosure as it would appear if the enclosure was sliced down the center as illustrated below.  

   

The springs could be replaced with magnets.  The force characteristics of magnets should remain constant with time.

The cylinder would be spun in a vacuum at a very high RPM but, as far as the interface/gap between the outer ring magnets and the iron piece are concerned, whether or not the cylinder is spinning is irrelevant.

The top drawing below shows the normal condition.

The middle drawing shows the situation when the cylinder comes closer to the assembly that holds the head.

The bottom drawing shows the situation when the cylinder is farther away from the assembly that holds the head.

 

 

   

The head is a dot.

What are shown as springs in the above drawings could be replaced with  permanent magnets as shown below.

 

  

 

The following drawing shows how a small dc current could be used to adjust, calibrate or zero the flying height.

 

 


The following drawing shows a possible way of creating an assembly that automatically flies the head and also allows for head movement (seeking).



 

Servo-Controlled Head to Surface Positioning

A low power laser could be used to measure the head-to-surface distance or the distance could possibly be determined by examining the signal from the head and a miniature actuator – like the larger ones used in disk drives to position the heads from track to track – could control the position of the head with respect to the surface as illustrated below.




Efficiency of Recording Objects

We define “efficiency” of a disk or cylinder to mean the number of square inches of recording surface area that will fit in a cubic inch of volume/space.

In other words, if a disk or drum contains 1 square inch of recording surface area and it will fit into a box that has 1 cubic inch of volume, then that recording object would have an efficiency of 1.  If a disk or drum contains 2 square inches of recording surface area and it fits into a box with 1 cubic inch of volume, then it would have an efficiency of 2, etc.

The efficiency of disk drives compared to cylinder drives is shown in the next section.


Comparing Disk Drives to Cylinder Drives

Let's compare a single-platter 10,000 RPM 300 GB Seagate Savvio 10K disk drive with 200,000-400,000 RPM cylinder drives.

The dimensions of the disk drive box are 0.591 in x 2.760 in x 3.955 in.

The volume of the disk drive box is Vd= 6.451 in3.

The number of square inches of recording surface are on the disk is approximately 2 x (p x 1.252p x .6252) = 7.36 in2

The diameter of the disk is 2.5 inches so the width of the disk drive box is approximately 0.25 inch wider than the diameter of the disk.

Consider a 0.5 inch diameter cylinder drive.  The cylinder will fit into a box that is 0.75 in x 0.75 in x L in.

What would L have to be so that the cylinder contains 7.36 in2 of recording surface area?

In that case, p x 0.5 x L = 7.36 in2      so L = 4.68 in.

If we allow 1 inch for bearings and other overhead the length will be around 6 in.

Each track on the cylinder drive is identical to every other track so most-likely significantly more bits could be stored per square inch on a cylinder than on a disk because a much more sophisticated channel equalization scheme could be implemented since the channel characteristics from track to track do not change.

The only way to really know what is achievable is to design and build a prototype cylinder drive.

The following Table compares a disk drive with several cylinder drives.

 Recording Object Disk CylinderCylinder Cylinder 
 Diameter 2.5 in0.5 in1.0 in 1.0 in 
 Square Inches of Recording Surface Area7.36 7.36 7.36 7.36 
 Length6 in 3 in3 in 
 Capacity300 GB300 GB 300 GB 300 GB 
 Average Latency3 msec 0.15 msec 0.15 msec 75 usec 
 RPM 10,000 200,000200,000 400,000 
 GB per cubic inch 46.5 88.9 64.064.0 
 Power Consumption ~7.5 W < 1W (est) < 1W (est) < 1W (est)

Someone could object that the above Table is not a fair comparison because up to 3 disks can be installed in a Savvio 10K drive to make it a 900 GB drive instead of a 300 GB drive consuming the same volume.  Someone else might argue that a cylinder drive where all the tracks are identical could hold 3x as many bits per square inch as a disk.  These types of arguments cannot be answered without actually designing, developing and building a prototype cylinder drive.  The point of the Table is to illustrate that if it is possible and practical to implement cylinder drives, they will probably be "roughly" equivalent to high-performance disk drives when it comes to number of GB/in3 if the diameters of the cylinders are in the range of around 0.5 to 1 inch but that, when it comes to performance and power consumption, cylinder drives as described on this web page would clearly be superior to disk drives.

In order to keep the power consumption at a reasonable level a Savvio 15K uses much smaller diameter disks and the capacity per disk is much less than for the Savvio 10K.  Conventional disk drives would not be able to compete with cylinder drives when it comes to power consumption and latency.  Disk drives have performance limitations that cannot be overcome.  If you spin a disk faster in air or helium it requires much more power so it is impossible to decrease latency and power simultaneously unless you also reduce capacity.  Replacing air with helium helps, but is nowhere near as good as a vacuum.

Appearance

Once a cylinder drive is fully assembled it would look as shown below.




This type of memory component would be used with a 2D-RS RAID error-correction system so even if the component is not ultra-reliable, it still could be used with a storage system that tolerates numerous component failures.  A failed unit is just removed and replaced like light bubs in a large department store.

Some might object about the shape of a cylinder drive and say the cylinder drive has an undesirable shape and that drives must be thin for laptop and notebook computers.

The answer to that objection is that there is an undeniable move to “the cloud” and the shape of drives in the cloud is irrelevant. What is relevant is the amount of space and power they consume and their performance and reliability.

Most-likely laptop and notebook computers will contain solid-state storage and large deposits of data will reside either in the cloud or in home servers where shape is also irrelevant.

Comments

Here's a link to an almost identical web page that has some comments on the bottom of the page.  Those comments are not on this page because that page was created and the comments made before this website was created.   If you would like to make comments on this page, e-mail me and I will enable you to post comments.


Ċ
Phil White,
Feb 27, 2014, 4:32 PM