Large flash chips are obtainable?

[ali_asadzadeh]

and how do you want to control those ? you are going to need a flash controller that can do the command queueing and wear leveling. That's not chicken shit to design ! and then there is the cache required

I have ideas regarding that part

[free_electron]

and how do you want to control those ? you are going to need a flash controller that can do the command queueing and wear leveling. That's not chicken shit to design ! and then there is the cache required

I have ideas regarding that part

ok. Nice. and how are you going to build that silicon ? Unless you have that part figured out , obtaining the flash chips is the least of your worries. I worked in datastorage for a large part of my life , designing IC's for drives. Plenty of controller companies came and went. sandforce and others. they either died because they were behind technologically, or were absorbed by the flash makers themselves. There is an intricate interplay between flash operation and the wear levelling. flash devices have command sets and depending on the data load the best commands need to be selected to optimize speed and reliability. That is a very complex matter. So the SSD makers use flash devices that offer a command set that works best with their controller. That's the reason that all the SSD makers are now owned by flash manufacturers and have the controller design in-house.

unless you know the instruction set of your particular device you won;t be able to build a controller or optimize it. and you can bet top dollar that information is locked behind a whole bunch of NDA's , if even available at all... these guys don;t want competition. they are already in a race to the bottom and their only diversifying aspect is the secret sauce : the interplay flash/controller. You try to upset that ? not going to work. if you got some breakthrough algorithm you can try selling that to them. but if you want to do the job of actually making drives .. go design your own flash.

[ali_asadzadeh]

ok. Nice. and how are you going to build that silicon ? Unless you have that part figured out , obtaining the flash chips is the least of your worries. I worked in datastorage for a large part of my life , designing IC's for drives. Plenty of controller companies came and went. sandforce and others. they either died because they were behind technologically, or were absorbed by the flash makers themselves. There is an intricate interplay between flash operation and the wear levelling. flash devices have command sets and depending on the data load the best commands need to be selected to optimize speed and reliability. That is a very complex matter. So the SSD makers use flash devices that offer a command set that works best with their controller. That's the reason that all the SSD makers are now owned by flash manufacturers and have the controller design in-house.

unless you know the instruction set of your particular device you won;t be able to build a controller or optimize it. and you can bet top dollar that information is locked behind a whole bunch of NDA's , if even available at all... these guys don;t want competition. they are already in a race to the bottom and their only diversifying aspect is the secret sauce : the interplay flash/controller. You try to upset that ? not going to work. if you got some breakthrough algorithm you can try selling that to them. but if you want to do the job of actually making drives .. go design your own flash.

Ok, fair enough, Try it on FPGA and sell the IP sounds good to me too.

[ejeffrey]

AFAIK, most high capacity flash chips use parallel interfaces.  Certainly that is what is used in consumer SSDs.  I don't know if you can find QSPI or other "serial" flash chips in such large capacity.  Why is that important for your application?

My Idea needs a serial Flash chip the problem is that I could not find large ones, even with Big prices to pay.

Ok if you don't explain what you are doing it's hard to help.

Probably your best bet is to get some appropriately sized SD cards.  AIUI they have parallel flash chips inside but with a controller that implements a couple variants of a serial interface. I await a cryptic comment about how that won't work for your application.

[CChin254]

That's possible to use FPGAs to control the SSD. Here is the block diagram of one of the worlds highest capacity SSDs.

[free_electron]

Ok, fair enough, Try it on FPGA and sell the IP sounds good to me too.

That could be done , but you will need one hell of an FPGA to create the flash controller. you will also need an external processor or four ..MAybe a Xilinx Zynq can do it. The flash controllers have some oomph behind them.
It may be easier to model this as opposed to try to build it in hardware. create the algorithm , model that. forget physical implementation. it will be very hard to get to the real speed.

[NiHaoMike]

That's possible to use FPGAs to control the SSD. Here is the block diagram of one of the worlds highest capacity SSDs.

All that capacity only to bottleneck it with a humble SATA connector. I would expect PCIe or something even more exotic for such a high end SSD.

[Berni]

That's possible to use FPGAs to control the SSD. Here is the block diagram of one of the worlds highest capacity SSDs.

All that capacity only to bottleneck it with a humble SATA connector. I would expect PCIe or something even more exotic for such a high end SSD.

That's the block diagram of one of these:
https://nimbusdata.com/products/exadrive/

The point if these SSDs is not speed but storage density. It can fit 100TB of storage in the size of a regular hard drive and so can fit into a single server hotswap bay. So with these you can easily get a petabyte of storage in a small server while making less noise, less heat, consuming less power and still being orders of magnitude faster than a hard drive when it comes to random access speeds. Stick a bunch of these on a fancy RAID controller and you still get blazing speeds in the end.

Admittedly it is a pretty niche product but there are some niche costumers that will pay up the big bucks for it.