Practical Chip Design

Jul 19 2008 1:21PM | Permalink | Email this | Comments (0) |
Blog This! using:  Blogger.com | LiveJournal |
Digg This | Slashdot This | add to Del.icio.us

There has been so much talk about lithography effects in metal on advanced processes that any time we hear terms such as OPC or sub-wavelength, we see images of pinched-off metal-1 traces. But the critical metal layers are not the only place things can go awry: there's also front-end-of-line processing. There are enormous struggles with contacts, for instance. But even before that, there is the matter of the diffusions that define the N and P regions of the silicon. Obviously if they are not in the right place, or not the right shape, bad things are to come.

But this raises an interesting question. If you have a lithography problem on a metal layer or a contact layer, you can pretty much just look at it with a scanning electron microscope and see what happened. The source of the problem may not be obvious, but the result will be. So what do you do to visualize diffusion problems? Doped silicon doesn't look all that different from undoped silicon.

One answer comes from a fascinating case of cross-fertilization at Agilent Technologies. Agilent has for some time been in the business of building atomic-force microscopes (AFMs.) These devices move a very tiny cantilever arm with an even tinier point on it across the surface of a sample, recording the topology of the surface by means of a laser rangefinder arrangement that measures the height of the tip above a reference plane. In addition to simply mapping topology, by changing the probe you can make many other interesting surface measurements, including atomic forces above the surface, heat, reactivity to specific antibodies—the list makes fascinating reading.

Thinking along these lines, somebody at Agilent came up with a great idea. Maybe you can't see dopant concentration from the surface of a silicon lattice. But the diffusion does change the electrical properties at the surface. Specifically, it changes the impedance to ground. So if you could connect an impedance bridge, or something of the sort, to the probe of an AFM, in principle you could map the surface of a chip and include the dopant concentration along with the topology.

Nice thought. But at the frequencies you'd need to detect such small changes in capacitance, connecting an impedance bridge through a highly flexible cantilever to a tip that's essentially a nanoparticle is a non-starter. That's where the cross-fertilization part comes in.

Another division of Agilent makes a totally unrelated product that happens to solve the problem: a Microwave Vector Network Analyzer. The analyzer measures impedance difference at a network node by generating a calibrated incident microwave, measuring either the transmitted or reflected microwave, and comparing the two. The important part here is that the instrument can make an absolute, not relative, measurement of capacitance or or dC/dV, by applying a signal and receiving a signal from a single point of contact. So all you need to do is make a very high-quality 50-Ohm connection from the VNA to the tip of the AFM probe, and you can get all the data you need to calculate dopant concentration in the silicon.

Amazingly enough, the microwave measurements team at Agilent was able to design such a connector, according to Agilent applications manager Jeff Jones. The connector joins the VNA and the microscope into a single instrument that can simultaneously map the topology and dopant concentration on the surface of a die with horizontal resolution on the order of tens of nanometers and capacitance resolution on the order of 1 atoFarad.

As a diagnostic tool this opens a huge window for visual inspection, Jones says. It is now possible, for example, to examine a failed SRAM cell at below the resolution of the lithography equipment, and to not only see the surface geometry—which might tell you if metal, contact, or dielectric material was missing—but also to image the dopant concentrations at the surface. That will tell you if a diffusion has gone bad. A whole new—and unfortunately not unimportant—range of failures has become visible.

Related entries in: Microscopy | Semiconductors |