Quote from: TyMoore on 05/22/2010 03:07 pmIt sounds as if off the shelf processors are O.K. as long as thorough error checking and correcting is implemented in the design. Perhaps a block of processors operating in a 5 way voting scheme--similar to the Space Shuttle GPC's--could be used for Criticality 1 applications. Data processing and non-critical applications could use more conventional networks...Error correcting and checking is not a kind of bolt-on, it needs to be inherent in the full design, top to bottom. Here's the list of rad-hardened CPU's on the market taken from wikipedia:BRE440, Proton200k, Proton 100k, RCA1802, System/4 Pi, RAD6000, RAD750, RH Pentium, RH32, RHPPC, SCS750, ERC32 SPARC, LEON SPARC, RH1750, Coldfire M5208, Mongoose-VOf these, the Protons and SCS750 do just this kind of "vote-tally" approach, but it needs to be done at the chip level. Implimenting it on the board level, using commodity PC CPU's would be far more inefficient, and expensive, than to use the rad-hardened designs already in use.
It sounds as if off the shelf processors are O.K. as long as thorough error checking and correcting is implemented in the design. Perhaps a block of processors operating in a 5 way voting scheme--similar to the Space Shuttle GPC's--could be used for Criticality 1 applications. Data processing and non-critical applications could use more conventional networks...
Quote from: Downix on 05/22/2010 06:26 pmQuote from: TyMoore on 05/22/2010 03:07 pmIt sounds as if off the shelf processors are O.K. as long as thorough error checking and correcting is implemented in the design. Perhaps a block of processors operating in a 5 way voting scheme--similar to the Space Shuttle GPC's--could be used for Criticality 1 applications. Data processing and non-critical applications could use more conventional networks...Error correcting and checking is not a kind of bolt-on, it needs to be inherent in the full design, top to bottom. Here's the list of rad-hardened CPU's on the market taken from wikipedia:BRE440, Proton200k, Proton 100k, RCA1802, System/4 Pi, RAD6000, RAD750, RH Pentium, RH32, RHPPC, SCS750, ERC32 SPARC, LEON SPARC, RH1750, Coldfire M5208, Mongoose-VOf these, the Protons and SCS750 do just this kind of "vote-tally" approach, but it needs to be done at the chip level. Implimenting it on the board level, using commodity PC CPU's would be far more inefficient, and expensive, than to use the rad-hardened designs already in use. True, but a board or chip utilizing ECC and vote-tallying could be manufactured using the exact same processes as are used for COTS components whereas rad-hard components use different manufacturing techniques, right?
Here's the list of rad-hardened CPU's on the market taken from wikipedia:BRE440, Proton200k, Proton 100k, RCA1802, System/4 Pi, RAD6000, RAD750, RH Pentium, RH32, RHPPC, SCS750, ERC32 SPARC, LEON SPARC, RH1750, Coldfire M5208, Mongoose-V
Quote from: Robotbeat on 05/22/2010 06:32 pmQuote from: Downix on 05/22/2010 06:26 pmQuote from: TyMoore on 05/22/2010 03:07 pmIt sounds as if off the shelf processors are O.K. as long as thorough error checking and correcting is implemented in the design. Perhaps a block of processors operating in a 5 way voting scheme--similar to the Space Shuttle GPC's--could be used for Criticality 1 applications. Data processing and non-critical applications could use more conventional networks...Error correcting and checking is not a kind of bolt-on, it needs to be inherent in the full design, top to bottom. Here's the list of rad-hardened CPU's on the market taken from wikipedia:BRE440, Proton200k, Proton 100k, RCA1802, System/4 Pi, RAD6000, RAD750, RH Pentium, RH32, RHPPC, SCS750, ERC32 SPARC, LEON SPARC, RH1750, Coldfire M5208, Mongoose-VOf these, the Protons and SCS750 do just this kind of "vote-tally" approach, but it needs to be done at the chip level. Implimenting it on the board level, using commodity PC CPU's would be far more inefficient, and expensive, than to use the rad-hardened designs already in use. True, but a board or chip utilizing ECC and vote-tallying could be manufactured using the exact same processes as are used for COTS components whereas rad-hard components use different manufacturing techniques, right?You are ignoring the volume issue. At the volume of manufacturing you're talking, under 100 units per year, the extra complexity will cost you more even with the cheaper components. The beauty of RAD-hardened systems is that you can use commodity manufacturing techniques, just using harder materials. I've worked with systems with volume of under 20 units a year, when discussing such low volume of work, the higher-quality of components always win out over cheaper.
Quote from: Downix on 05/22/2010 06:44 pmQuote from: Robotbeat on 05/22/2010 06:32 pmQuote from: Downix on 05/22/2010 06:26 pmQuote from: TyMoore on 05/22/2010 03:07 pmIt sounds as if off the shelf processors are O.K. as long as thorough error checking and correcting is implemented in the design. Perhaps a block of processors operating in a 5 way voting scheme--similar to the Space Shuttle GPC's--could be used for Criticality 1 applications. Data processing and non-critical applications could use more conventional networks...Error correcting and checking is not a kind of bolt-on, it needs to be inherent in the full design, top to bottom. Here's the list of rad-hardened CPU's on the market taken from wikipedia:BRE440, Proton200k, Proton 100k, RCA1802, System/4 Pi, RAD6000, RAD750, RH Pentium, RH32, RHPPC, SCS750, ERC32 SPARC, LEON SPARC, RH1750, Coldfire M5208, Mongoose-VOf these, the Protons and SCS750 do just this kind of "vote-tally" approach, but it needs to be done at the chip level. Implimenting it on the board level, using commodity PC CPU's would be far more inefficient, and expensive, than to use the rad-hardened designs already in use. True, but a board or chip utilizing ECC and vote-tallying could be manufactured using the exact same processes as are used for COTS components whereas rad-hard components use different manufacturing techniques, right?You are ignoring the volume issue. At the volume of manufacturing you're talking, under 100 units per year, the extra complexity will cost you more even with the cheaper components. The beauty of RAD-hardened systems is that you can use commodity manufacturing techniques, just using harder materials. I've worked with systems with volume of under 20 units a year, when discussing such low volume of work, the higher-quality of components always win out over cheaper.Well, I certainly yield to your greater experience.I agree that you should use rad-hardened components if they are available and meet your requirements! In that case, rad-hard components would be off-the-shelf. If they don't meet your requirements, then perhaps it would be a good idea to do these more complicated end-to-end ecc (and voting) methods versus spending a bunch of money to enhance the state-of-the-art in rad-hard fabrication methods.
in those cases it is cheaper to actually roll your own with rad-hardened fpga technology or ASIC.
Quote from: Downix on 05/23/2010 01:02 amin those cases it is cheaper to actually roll your own with rad-hardened fpga technology or ASIC. Rad-hard FPGAs seem to have a lot of potential for one-off science missions, but last I heard, they're still pretty low in transistor count. Has this been getting better?
Quote from: simonbp on 05/23/2010 05:08 amQuote from: Downix on 05/23/2010 01:02 amin those cases it is cheaper to actually roll your own with rad-hardened fpga technology or ASIC. Rad-hard FPGAs seem to have a lot of potential for one-off science missions, but last I heard, they're still pretty low in transistor count. Has this been getting better?FPGAs follow Moore's Law. Due to the high routing and programming overhead (several transistors for each of the user's gates) they are at least a decade behind dedicated chips.At four launches a year LV are not exactly a mass production item, making using FPGA in them viable.
O.K., I've got a question:Without getting into too many specific, proprietary, or classified details; how much (in a rough ball park) does it cost to develop a new Rad Hard FPGA processor?
How many ARM7 cores can dance on the head of a rad-hard FPGA pin?Edit: Oops, apparently I meant CortexM1 not ARM7.http://www.actel.com/products/mpu/CortexM1/
Usually the grades of ICs (commercial, industrial, military) just denote the level of testing that they have undergone, IE, room, temp, extended temp. Tri-temp testing can cost as much as the die and package, and yield loss usually goes up a couple percent, justifying the higher prices.There are plenty of COTS rad-hard parts available. I think COTS usually means 'already available at a reasonable price' as opposed to building a custom widget for your application or paying rapacious prices for it. Pre-existing products have also usually had time to build a track record of success, whereas something new is more likely to have bugs left in it.
Quote from: josh_simonson on 05/25/2010 11:04 pmUsually the grades of ICs (commercial, industrial, military) just denote the level of testing that they have undergone, IE, room, temp, extended temp. Tri-temp testing can cost as much as the die and package, and yield loss usually goes up a couple percent, justifying the higher prices.There are plenty of COTS rad-hard parts available. I think COTS usually means 'already available at a reasonable price' as opposed to building a custom widget for your application or paying rapacious prices for it. Pre-existing products have also usually had time to build a track record of success, whereas something new is more likely to have bugs left in it.If there are lots of COTS rad-hard parts, they should be used!!! In that case, using non-rad-hard parts is not going to make a lot of sense.What I'm really interested in is high-temp and low-temp parts so that thermal management requirements become extremely relaxed.
This thread seems sort of related to a topic I want to ask about, so I'm bumping it rather than creating my own.I can see why commercially available electronics are unsuited for control, avionics, ect, but are they usable in non-critical areas? Would one be able to fly an iPod or Kindle, or would one need to develop a space-rated alternative?
Also, I can get why processors and chipsets have to be shielded, but what about things like LCDs? Do those also require special versions?