Shop OBEX P1 Docs P2 Docs Learn Events
P2 Flash - Is it possible with On Semi ? — Parallax Forums

P2 Flash - Is it possible with On Semi ?

Cluso99Cluso99 Posts: 18,069
edited 2013-11-15 23:53 in Propeller 2
I noticed that On Semi make Flash Memory (they acknowledge a license to another manufacturer).

I wondered if this process is available to Parallax and if so, would there be enough room on the die to add a small amount (say 4KB+) for boot code?

The boot code could then either contain the monitor or boot from SPI Flash or SD or whatever.

There are probably many reasons not to do this, in particular delays, but thought it was worth asking anyway.

Comments

  • jmgjmg Posts: 15,173
    edited 2013-11-12 14:12
    Flash needs more process steps, and cannot usually read at the same speed, so has to be a separate cell.
    For a tiny amount, it is not worth it.

    I think OTP is more possible/practical, ( see XMOS), but that is also a separate cell.

    The trade off is more cells is more die area, and more to go wrong.

    The Present boot rom, is hand-stuck RAM cells, which is compact and fast and simple.

    Downside, is the manual patching, and so changes are (I think) not even top-metal only as most ROM is, but need a multiple layer re-spin.
    That means checking ROM is going to be very important.
  • Beau SchwabeBeau Schwabe Posts: 6,566
    edited 2013-11-12 14:58
    "Downside, is the manual patching, and so changes are (I think) not even top-metal only as most ROM is, but need a multiple layer re-spin.
    That means checking ROM is going to be very important."
    - It doesn't really matter what layer it is, you still have to fabricate ALL layers. We have however created our ROM cell, so that it can be set to a "1" or cleared to a "0" depending on the position of a contact .... or single layer ... which means that only one mask needs to be changed. The ROM bits (contacts) are also placed with a script, so any changes can be automated in just a couple of seconds.

    As far as Flash or OTP, your right, that means more layers, and more layers equals more $$$. Additionally, the current process that we have targeted does not support Flash without significant changes.
  • jmgjmg Posts: 15,173
    edited 2013-11-12 15:13
    It doesn't really matter what layer it is, you still have to fabricate ALL layers. We have however created our ROM cell, so that it can be set to a "1" or cleared to a "0" depending on the position of a contact .... or single layer ... which means that only one mask needs to be changed.

    In classic ROM parts, that was a top layer. Which layer is it in the Prop 2

    Short term it does not matter much, but longer term you may find volume customers, willing to pay the mask prices.
    The ROM bits (contacts) are also placed with a script, so any changes can be automated in just a couple of seconds.

    I guess that works for any size ROM ?

    Is that flow die-proven / tested yet - or did the last pass not allow that level of testing ?
  • Beau SchwabeBeau Schwabe Posts: 6,566
    edited 2013-11-12 15:48
    "In classic ROM parts, that was a top layer. Which layer is it in the Prop 2" ... it's literally the contact layer, which allows connection to diffusion,poly, and M1. Top layer is also a relative term. A ROM cell by itself or RAM cell for that matter should not go above M1. However in the hierarchical connections our process goes to M6. And while it's true that you don't have routing over Memories, you still have density metal that needs to be placed which does go over memories.


    "Is that flow die-proven / tested yet - or did the last pass not allow that level of testing ?" - A similar script was used for an earlier test die and yes that proved to be correct. The same script was used for that as was in P2, but the pitch and step positions were adjusted accordingly.
  • Cluso99Cluso99 Posts: 18,069
    edited 2013-11-12 17:27
    As far as Flash or OTP, your right, that means more layers, and more layers equals more $$$. Additionally, the current process that we have targeted does not support Flash without significant changes.
    Thanks Beau. Just thought I would ask in case the On Semi process was different.
  • evanhevanh Posts: 15,913
    edited 2013-11-13 02:00
    Anyone know any details of Motorola's spin-offs? Google bought up the remaining consumer husk but what parts do OnSemi, Freescale and Everspin divide up into? Freescale seems to concentrate on uControllers (Does it have it's own fab?). Everspin is focused on MRAM but does it do anything else? OnSemi, obviously punches out a bunch of generic parts, but presumably is into contract fab'ing these days, right?
  • jmgjmg Posts: 15,173
    edited 2013-11-13 11:46
    evanh wrote: »
    OnSemi, obviously punches out a bunch of generic parts, but presumably is into contract fab'ing these days, right?

    See
    http://www.onsemi.com/PowerSolutions/content.do?id=16558

    Some interesting split voltages shown in the tables there.

    http://www.onsemi.com/PowerSolutions/content.do?id=16617
    http://www.onsemi.com/PowerSolutions/content.do?id=16622

    OnSemi also has the ex-Sanyo Microcontroller & other semi lines
    http://www.onsemi.com/PowerSolutions/taxonomy.do?id=101533
  • Beau SchwabeBeau Schwabe Posts: 6,566
    edited 2013-11-13 12:33
    Keep in mind also with large companies that you are basically a resource. When I was at National Semiconductor I was contracted within National to do layout work for Hewlett Packard, Motorola, Qualcomm, and IBM.... so many of the services/options you see within what seems to be one company could span across several sources.
  • Erik FriesenErik Friesen Posts: 1,071
    edited 2013-11-15 03:48
    I wondered if this process is available to Parallax and if so, would there be enough room on the die to add a small amount (say 512KB+) for code? :-)
  • Mike GreenMike Green Posts: 23,101
    edited 2013-11-15 08:08
    512KB is not a small amount and the die is already quite large for a microcontroller.
  • evanhevanh Posts: 15,913
    edited 2013-11-15 13:44
    As well as space considerations it has been repeatedly said that, a) the chosen process doesn't support Flash, and, b) on the processes that do support Flash it involves more layers which is more costly. The process is available, but at a cost.

    The process points also apply to MRAM, which, imho, is by far the superior option over Flash since it would fully replace HubRAM but at a higher density.

    Multi-ported MRAM anyone? Build CogRAM out of it too ...
  • Erik FriesenErik Friesen Posts: 1,071
    edited 2013-11-15 13:47
    I know its not supported, I have no illusions, just thought since cluso asked for 4kb...
  • evanhevanh Posts: 15,913
    edited 2013-11-15 14:08
    512kB MRAM would likely be achievable in without any increase in die size.
  • jmgjmg Posts: 15,173
    edited 2013-11-15 15:08
    evanh wrote: »
    512kB MRAM would likely be achievable in without any increase in die size.

    ?? You mean Magnetic memory ? - that needs extra fab steps and it has speed and lifecycle issues.

    High MHz is going to be one of the important leverage points of P2, and that excludes FLASH.

    SRAM is really the only choice, cheap and fast and proven.
  • ozpropdevozpropdev Posts: 2,792
    edited 2013-11-15 15:28
    What about a separate flash die that's included at the packaging stage?
  • jmgjmg Posts: 15,173
    edited 2013-11-15 16:15
    ozpropdev wrote: »
    What about a separate flash die that's included at the packaging stage?

    Other companies do that, but their flash size decisions tend to be defined for them (eg FPGA).

    In Prop 2, it gets hard to compete in price on the smallest Flash parts, and for larger flash, finding room and selecting a size make it hard to hit critical mass in a product-line basis.

    Because RAM is so valuable (/costly), during discussion on ROM size, there was talk of a separate boot-device, but you still need some tiny ROM to talk to that, so the final choice of 'smallest practical' ROM is best.

    Chip seemed to always pack more into less space, and I'm not sure of the final ROM size post-threading.

    It does mean Fonts and Spin have to go from the ROM, but they can be more compile-time managed.

    I see some long term potential with large customers in the Prop 2 ROM flows, but first get a std part working....
  • evanhevanh Posts: 15,913
    edited 2013-11-15 16:38
    jmg wrote: »
    ?? You mean Magnetic memory ? - that needs extra fab steps and it has speed and lifecycle issues.
    I wasn't advocated MRAM over SRAM. I was saying that if MRAM were to be used it would indeed, unlike Flash, directly replace the SRAM. However, it's a whole other process according to Beau and Chip.
    High MHz is going to be one of the important leverage points of P2, and that excludes FLASH.
    But doesn't exclude MRAM.
    SRAM is really the only choice, cheap and fast and proven.
    MRAM is a good choice for many reasons but it's also more expensive for the time being.
  • evanhevanh Posts: 15,913
    edited 2013-11-15 16:49
    Life cycles meaning write cycles I presume? MRAM is not limited at all. Flash/EEPROM is limited on write cycles. FRAM is limited on both read and write cycles.
  • jmgjmg Posts: 15,173
    edited 2013-11-15 17:23
    evanh wrote: »
    MRAM is not limited at all.

    Which MRAM are you talking about ?

    I see Everspin give a lower clock speed for read, unlike most other technologies, maybe because read needs the sense-amps ?

    In their parallel series there is nothing showing under 35ns.
  • evanhevanh Posts: 15,913
    edited 2013-11-15 18:33
    jmg wrote: »
    In their parallel series there is nothing showing under 35ns.

    I was of course talking about number of writes (wearing the part out), nothing at all to do with access times. That said, while I can't say for sure how fast MRAM would be as a process replacement for HubRAM your reference is like comparing external SRAM chips with internal caching, not quite apples with apples.

    The claims are that MRAM is close to SRAM access speeds on a one-to-one footing. Most super fast SRAM these days will be using much finer processes than 180 nm. What is the expected access times for the Prop2 HubRAM SRAM cells?
  • evanhevanh Posts: 15,913
    edited 2013-11-15 18:37
    And the only reply I've seen from Beau/Chip on using MRAM for HubRAM was one of cost (In change of process and using more layers), not performance.
  • jmgjmg Posts: 15,173
    edited 2013-11-15 19:24
    What is the expected access times for the Prop2 HubRAM SRAM cells?
    To avoid any talk of wait-states, it needs to be <5ns cycle times to allow adjacent 200MHz COG slots.

    MRAM is academic, is it too expensive, so no one checked the speed, but that looks to be too slow as well.
    Given it has to read a magnetic state, it is no real surprise it is slower. (cheap SRAM comes down to 8ns, async)

    Either shortcoming is enough to kill it, it does not take both.
  • evanhevanh Posts: 15,913
    edited 2013-11-15 19:32
    Of course it's academic but Flash is not even an option as an SRAM/DRAM replacement. This has always been a comparison to Flash.
  • Cluso99Cluso99 Posts: 18,069
    edited 2013-11-15 21:53
    Sorry I opened a can-of-worms. I just thought with the change to On Semi it was worth asking, and that was answered quite well (as a no) by Beau.

    What Chip & Beau have done for the small ROM is to effectively hardwire the SRAM, so the SRAM layout done only had minimal changes. This is again, quite smart.
    For anyone interested, IIRC it is in the main Prop thread.

    Can we close this discussion now.
  • evanhevanh Posts: 15,913
    edited 2013-11-15 23:53
    Don't know if I'd call it a can of worms, that might come in the future. Parts are speculative and there never was any chance of it applying to the P2 anyway. I think the topic was worth while just for awareness.
Sign In or Register to comment.