> On Dec 17, 2018, at 12:51 PM, Guy Sotomayor Jr via cctalk 
> <cctalk@classiccmp.org> wrote:
> 
> 
> 
>> On Dec 16, 2018, at 10:40 PM, Chuck Guzis via cctalk <cctalk@classiccmp.org> 
>> wrote:
>> 
>> On 12/16/18 11:21 AM, Paul Koning wrote:
>> 
>>> If you simply want non-volatile memory, the obvious answer is SRAM with 
>>> battery backup and a small FPGA to do the interfacing.
>> 
>> I proposed nvRAM - CMOS SRAM backed by cell-for-cell flash.  Loads SRAM
>> from flash on power-up and stores into flash at power-down.  All that's
>> needed is a capacitor to extend the power-down cycle a bit.
>> 
>> Very fast, available in 8 to 32-bit wide architectures, up to 16Mbit per
>> package.
>> 
>> Claims to be guaranteed for 1M power cycles and doesn't require a battery.
> 
> Except it is *much* more expensive than MRAM.  32x8 NVSRAM is $18.50 in qty 1 
> from Digikey.
> A 64Kx16 MRAM is $11.84 in qty 1 from Digikey.  MRAM requires no additional 
> circuitry so that
> also reduces the overall cost (and has unlimited write endurance).
> 
> If it sounds like I’m harping on MRAM, maybe I am.  I’ve looked at the 
> various technologies in
> detail (what’s available, cost, interfacing, etc, etc) for years and for 
> anything that requires
> non-volatility, MRAM wins until you get into seriously large sizes at which 
> point you need to
> go to FLASH for economics.

I'll go along with that.  I worked on a storage product that needed 
non-volatile memory in modest sizes for tracking transaction states.  The first 
product used FRAM for that, the second used SRAM backed by a small battery 
("coin cell") and the third used MRAM.  All worked fine.  FRAM supposedly has 
endurance limits but they are high enough they weren't a concern.  The main 
issue was the size limits, at least at the time (2003-ish).

For simplicity, battery backed SRAM should be just as good as SRAM.  In any 
case, you presumably will need a CPLD or small FPGA for the interface protocol 
conversion.

        paul

Reply via email to