Each foundry has process nodes defined by the minimum size of key features. Recent parts are made in 90-nanometer processes and older parts in 180-, 130- and 110-nm processes. Foundries are now working on 65- and 45-nm processes. Each will be able to produce parts using several process nodes. Key features that must be examined to identify a device will vary between the different process nodes of the same manufacturer. To complicate things, devices from different foundries made at the same process node are very similar for competitive reasons: The fabless chip vendors who are the foundry's customers will take their business to the foundry with the best process.
Foundry ID is done in stages. The first determines the device's process node. The second identifies specific features to see which foundries have a process at that node and would likely have made the part. Finally, the results of the first two stages are compared with samples from known sources.
To determine the node at which the device was made, it must be de-encapsulated and cut to obtain a cross section. A scanning electron microscope photographs key features like the width and shape of the transistors and minimum metal line width and spacing. If possible, these measurements are compared with the design rules published by the foundries. With that information, the device's process node can be determined. Foundries that do not produce parts at this node can then be ruled out.
The next step is more difficult. It requires vetting the list of foundries with processes at this lithography, then picking the one that is the best match for the device. This step analyzes both noncritical features and the material in the different layers that make up the device.
Noncritical features are examined because they tend to differ between foundries. Where critical features are highly optimized for speed or low power and tend to be similar between the foundries at the same process node, noncritical features, including dummy features and the die edge seal, do not affect device performance and tend to differ. Similarly, test structures at the die's edges as well as alignment marks for the various masks will differ among foundries.
Material analysis looks for small differences, foundry to foundry, in the chemical composition of the layers that make up the chip. Differences will exist. For example, in doping low-k dielectric materials using both fluorine and chlorine, a foundry may introduce a particular material in every process from 130 nm and below. A competitor, however, may not have done the same until it develops a 110-nm process. These differences can distinguish between devices fabbed at different foundries.
Individual dice are laid out on a wafer in a checkerboard pattern. While the chip designer defines the area within each die, the area between the dice belongs to the foundry--real estate it uses for alignment marks and test structures. When the wafer is cut into individual dies, part of the "scribe channel" area between the dies is left. Using those pieces for foundry ID is complicated since, to some extent, they depend on the foundry's manufacturing and test equipment.
The final step compares these results to devices known to have been made by a specific foundry. To some extent this is difficult since, ideally, samples from each process node from all possible foundries are compared, complicating the probe. In the end, the results of a foundry ID will not prove that a chip was made at a specific foundry, just suggest a high probability that it was.
Armed with this information, the manufacturer must then ask the foundry to reveal who commissioned the devices. The foundry may or may not divulge this information.
It takes persistence to follow the chain to the end. The manufacturer must then determine the damages, to sales and brand name, to determine how far it will pursue the matter.
By David Fraser (davidf@semiconductor.
com), principal for IPinsights at Semiconductor Insights (Kanata, Ontario).
See related chart
See related chart