Slashdot Mirror
DARPA Looks To End the Scourge of Counterfeit Computer Gear
coondoggie writes "Few things can mess up a highly technical system and threaten lives like a counterfeit electronic component, yet the use of such bogus gear is said to be widespread. A new Defense Advanced Research Projects Agency (DARPA) program will target these phony products and develop a tool to 'verify, without disrupting or harming the system, the trustworthiness of a protected electronic component.'"
Why did "remote kill switch" and "built in spying" just pop into my head?
No, I don't remember your name. But the memory mapped screen on a TRS80 from 1977 is from 15360 to 16383 if that helps.
"SHIELD demands a tool that costs less than a penny per unit, yet makes counterfeiting too expensive and technically difficult to do"
and at the same time
"What SHIELD is seeking is a very advanced piece of hardware that will offer an on-demand authentication method never before available to the supply chain"
These appear to be mutually exclusive.
William George
Is this actually a proposal to provide a general solution to the halting problem for a potentially unpredictable(if parts of it are hidden by the bugged component) program running on logic that may deviate from expected behavior under unknown conditions, or is there some trick that makes it less hopeless?
Well when you read the article its just comes down to a glorified Certificate of Authenticity:
DARPA said it eversions this dielet will be inserted into the electronic component's package at the manufacturing site or affixed to existing trusted components, without any alteration of the host component's design or reliability. There is no electrical connection between the dielet and the host component.
So yeah, the first thing that will be counterfeited will be these dielets.
But even baring that, since it has no connection to the actual electronics and firmware, simply seeing it on the package means nothing if the part you are using was compromised before it came out of manufacture, or passed through hands with the capability to compromise it before it hit your loading dock.
I suppose this solves the problem of those clone factories that manage to completely clone a functioning board or component to the point that it actually operates (at some level) and inserts these into the supply chain as a money grab. It probably doesn't save you from espionage on a grander level such as state sponsored.
Sig Battery depleted. Reverting to safe mode.
Even without any counterfeiting, it sure would be a bummer if your 'verified +++ Secure!' chip fell victim to some subtle modification. Possibly even one that could be implemented by modifying the die after fabrication. Wouldn't be trivial; but if you only need to change a few hundred transistors and possess sufficient hardware and motivation...
It occurred quite a long time ago, but at the time no solution was proposed.
Regular steel bolts have hexagonal heads that are flat on top. Bolts made of high-strength steel are marked with three - if I recall correctly - radial lines.
You can see that it would be easy and cheap to mark a regular steel bolt with those three lines, then sell it for the high-strength premium.
This caused at least on death: a worker who was torquing a bolt while building the first Saturn car factory snapped the head off a bolt and fell to his death.
An Army general commented that when he took his battalions tanks out for training in the desert, their tracks were littered with bits of broken off bolts, as well as the occasional tank tread.
What they actually did about this was to test samples of bolt shipments, but such testing was very expensive and so could not provide good coverage.
However it has been years since I last heard about it. Has the counterfeit bolt problem been solved? If so how?
Please mail me URLs of software employers.
How can one discern between counterfeit and real, when both are coming off the same assembly line in China?
This is what is called "third shift" products, where the first two shifts make XYZ product for ABC corp, and the third shift makes XYZ Counterfeit for black market.
Agent K: A *person* is smart. People are dumb, stupid, panicky animals, and you know it.
Let us not forget the very real problem with NSA practices in our field, and suddenly why your technical computing cluster stopped working, or other equipment screwed up. Our friendly hackers at NSA and any bios changed while providing snooping, changing timing on cpu, etc.
I already had problems with my land line which a firmly believe was because the NSA and small local phone company kept having technical issues leaving with out phone service for extended periods with what I will assume was interface problems with the regular phone equipment and the NSA equipment locked away. The problems started each after making email contact with Congressional Reps. Sometime later on, Sen. Sanders (VT) asked the question about spying on congress, which you should already know about.
After the 3rd time started I decided a cell was the option, though I hate the quality. But at least the damn this would dial 911 if I happen to need it, and I can get voice mail and make calls.. you know, what was working fine before on the better system.
All before Snowden NSA reveal, so my complaints were brushed away at the time. although many of us here who follow technical and networking hires/info had clear signs this was occurring long before Snowden came on scene.
So, let's all not worry about those NSA ass-holes, you got nothing to hide, right? Why object otherwise.
It seems to me that most of you didn't bother to read the article. In a nutshell, DARPA wants a small electrically isolated chip that acts as a RFID chip and sends an encrypted response to an interrogation. Method of use
1. Specialized probe scans chip. Obtains serial number of chip.
2. Specialized probe sends serial number information to centralized server.
3. Centralized server sends back to probe query string.
4. Probe passes onto chip, the query string.
5. Chip sends back encrypted response to query string.
6. Probe passes back to centralized server, encrypted chip response.
7. Centralized server sends back to probe "good" or "bad" results.
Notice that the encryption key may be unique for each chip. The keys are known by the centralized server, but don't need to be known by anything else.
In order to create a counterfeit, the attacker needs to do one of two things.
1. Duplicate an existing chip to include the serial number and encryption key.
2. Create a new chip with a new serial number and encryption key and implant that serial number and key into the database maintained by the centralized server.
If an attacker is capable of compromising the central server, then it's game over. But the assumption is that is a "hard task". So the security is likely to be aimed at protecting the encryption key for each chip. Perhaps store the key in TLC Nand and arrange for the value to be corrupted if it's exposed to light (and of course, encapsulate the chip in an opaque material).
So when you manufacture a "non-counterfeit" component, you
1. Manufacture component.
2. Glue a chip to the component.
3. Register the chip with the centralized server.
To verify that a component isn't a counterfeit.
1. Scan for chip and do the entire song and dance to verify the chip.
Increased awareness, new federal laws holding manufacturers responsible for meeting the marking of their items, etc. I suspect it's still a problem though, just not as widespread as it once was. There is a good pdf guide from the gov on counterfeit bolts and other hardware items (I love the completely cosmetic gas cutoff valve on pg. 32)
I browse on +1 so AC's need not respond, I won't see it.
Used to be easy back when ICs had few layers and feature sizes that were resolvable with optical light microscopy.
I used to verify ICs for trustable computing back in the day, We would take aligned dieshots, develop them, project them through a red filter, project the master copy of the dieshot through a blue filter, overlay them and look for any large red or blue bits indicative that the metal was not the same. In light of recently published dopant sabotaged parts, it is obvious the technique we used back then was flawed, and not really applicable to modern chips which have many metal layers.
Destructive testing of representative samples can yield verification of all metal layers, but still doesn't cover the dopant sabotage technique (which we were not aware of at the time), of course you could try and slip things through by only sabotaging 5% or 10% of the parts.
I think a much more prevalent problem is counterfeit parts for commercial gain than sabotage, for example taking some cheap MOSFET with similar but worse characteristics and relabelling it as some expensive MOSFET. This happens frequently (found a batch of fake BJTs when I was building an amplifier, as the fake part had the wrong CBE pinout). Another common technique is taking low speed-grade DRAM, assembling it on DIMMs and programming SPD data that claims to be highspeed DRAM, often they don't even bother to change the labels on the device packages, as they are covered by a heatsink on most modern DIMMs.
The problem with out-of-grade counterfeiting is that the different grades are produced from identical masks, and can only be differentiated by very careful measurement of the device parameters. In some cases the counterfeits even meet every parameter, as they are produced by binning in the same way the the original manufacturer bins the parts prior to labelling. There are opamps which cost $40/each, and they are binned from the same line which makes parts costing $1/each, the manufacturer is even up front about this. Sometimes, the line produces more parts that would qualify for the $40/each part number than there is demand for the $40/each part, so the manufacturer just bins them as a lower price part. If the counterfeiter was upfront about it like the OEM, they would relabel them as their own part, with some guarantee about performance like the OEM provides, but then they wouldn't be a counterfeiter, but a legitimate re-binning service.
An example of legitimate rebinning is many of the high end audio equipment or lab equipment manufacturers. They often use commercial grade parts, have an internal test jig, and resell or dispose of parts that don't meet their higher-than-spec required parameters.