segunda-feira, 18 de abril de 2022

A project to increase the throughput of HDDs without increasing the cost of production.

A project to increase the throughput of HDDs without increasing the cost of production.

HDDs, on low-end (a 2TB HDD has a price identical to a 256GB SSD, 40€), have a price 8 times smaller than a SSD but have 30 times less IOPS and 5 times less throughput. As the reading/writing is done on only one side of a platter (1 bit HDD) and it is not possible to increase the rotation speed of the disk or the areal density, bigger HDDs are becoming even slower to backup, for example, for a speed of 100MBps (7200 rpm HDD) it takes 52 hours to backup a disk of 18TB.

To compete with SSDs on server market, HDDs have to increase their capacity to 100 TB by 2030 which, given the difficulty in increasing density, only is possible by increasing the number of platters (or the diameter of each platter). Then, even assuming an expensive 10000 rpm HDD (150MBps), the backup of 100TB HDD will take 200 hours.
Intel Mach 2, by using two independent actuatores, can double the throughput (no so strainforward!) but increases the complexity and price of the HDD (and do not increaseas IOPS).

Imagine that it is necessary to reset the system.
An organization has 10,000 customers and each uses 100GB, for a total of 1000TB.
Then, it uses 2 backup copies on HDD, (a RAID 99 + 1 file servers, 5400 rpm).
In the event of an attack, it will be necessary to restore the system that will take weeks.

Data must be organized "vertically".
My idea is to organize the data “vertically” with all heads being read/written simultaneously with just one actuator. In this way, keeping all the actual mechanics, velocity and density (and the cost, just need to change the data controller), it will be possible to increase the writing and reading throughput proportionally to the number of platters (although not increasing the IOPS).
One could have: 
                                                           7200RPM  / 5400RPM 
     2 bits HDD = 1 platter  -    2TB - 200MBPS  - 150MBPS
     4 bits HDD = 2 platters -   4TB - 400MBPS   - 300MBPS
     8 bits HDD = 4 platters -   8TB - 800MBPS   - 600MBPS
   16 bits HDD = 8 platters - 16TB - 1600MBPS - 1200MBPS

In the future, assuming a 50% increase in density (pitch from 25nm to 20 nm), one can have: 
   32  bits HDD = 16 platters -   50 TB - 3200MBPS - 2400MBPS
   64  bits HDD = 32 platters - 100 TB - 6400MBPS - 4800MBPS
Increasing HDD to 16/32 platters will increase thickness from 26 mm to 40/72 mm.

With the actual density, the first HDD (the 1956 IBM 350 HDD 24 inch with 50 platters) would have 580TB.

Fig. 1 - Schematic of a 16-bits HDD where data is read/writing simultaneously by all 16 heads moving together (i.e., in the same vertical position).

Fig. 2 - To double IOPS with two actuatores, they must be in opposite sides of the platter (wright).

A project to significantly lower the cost of producing of Intel Optane memory chips by verticalization.

(Probably this will not work because or the need for heating!)

Optane memory tecnology has potencial (it is RAM and 10 times faster than NAND memory) but it is is slowly fading out due to high production costs. Each layer require deposition, lithography and etching that is expensive.

My project is transforming Optane from planar to vertical, having just two lithography and etching fases, first, to etch holes (Fig. 1, Section view - 3) and, second, to etch trenches (Fig. 1, Section view - 6). 

The design woud be similar to 3D-NAND (inexpensive).

Fig. 3 - V-Optane construction operations (part 1)

Fig. 4 - V-Optane construction operations (part 2)

Fig. 5 - V-Optane top view

Fig. 6 - Reading/programming a bit

0 comentários:

Enviar um comentário

Twitter Delicious Facebook Digg Stumbleupon Favorites More

Design by Free WordPress Themes | Bloggerized by Lasantha - Premium Blogger Themes | Best Hostgator Coupon Code