We’ve been hearing and reading a lot about work being done at Georgia Tech’s PRC in developing glass interposer technologies as a low-cost alternative to silicon in 2.D and 3D applications. Mostly, it’s the cost benefits that are being touted. However, what sort of journalist would I be if I didn’t also report on the flip side of this technology? Today I spoke with Madhavan Swaminathan, who works with PRC and is the Director of the Interconnect and Packaging Center. He has been conducting research on both glass and silicon as the substrate material for interposers. His work is showing that silicon can outperform glass, particularly for high frequency applications. Swaminathan recently presented his work on this topic at Design Con, and he briefed me on his findings.
According to Swaminathan, crosstalk, insertion loss, power distribution, impedance, and thermal management are all key elements to consider when determining if Si or glass is a better material for TSV interposers in high frequency applications. Through electrical design and modeling methods, he put glass and Si through their paces for each element to determine the better performer.
Crosstalk
Due to the proximity of TSVs, crosstalk has become an issue, particularly when using CMOS grade silicon vs. the more expensive high grade (high resistivity) silicon. Shielding with ground vias around the signal via helps to eliminate crosstalk, but its not enough. Crosstalk is not a problem when using glass as the substrate material. Score one for glass!
Figure 1. Signal and ground TSV pair.
Insertion Loss
Glass is also a good dieletric material, noted Swaminathan, and insertion lossassociated with glass is considerably lower than with Si. The resistivity of glass is also high, which is critical. In fact, when comparing the coupling and insertion loss between TSVs and through glass vias (TGVs) a 3x improvement is achieved when using glass. Glass is also a better insulator than silicon. Score 2 for glass!
Impedance
This is a biggie. With regard to the power and ground plane, peak resonance in glass interposers is high. The larger the peak value the more difficult the power delivery becomes, because more noise is generated in the power distribution. The impedance is 10x lower for silicon as compared to glass, leading to better eye diagrams (Figure 2). It is Swaminathan’s determination that the low impedence is critical to the extent of cancelling out the glass’s superior performance in both the crosstalk and insertion loss areas. “Silicon is lossy and has more crosstalk,” he explained, “ but if you’re able to design the structures correctly where losses of silicon can be used to your advantage, than you can get better performance with silicon than you can with glass.”
Figure 2. These simulated eye diagrams at 3.2Gbps for glass (a) and silicon (b) show that jitter and eye opening are considerably improved in the silicon interposer. “
Thermal Management
Thermal management is another win area for Si. Silicon has better thermal conductivity than glass because heat spreads better through Si. Additionally, as the silicon interposer heats up it behaves better; demonstrating a 25% reduction in crosstalk. Particularly in high frequency applications, a silicon interposer outperforms its glass counterpart.
Summing it Up
Let’s first look at Si. According to Swaminathan, while insertion loss is high, there are workarounds. Crosstalk can be a killer. Accounting for temperature effects is critical because Si spreads heat and the increase in temperature reduces crosstalk. Si also provides advantages for power delivery.
Then there’s glass. While insertion loss and crosstalk is low, Swaminathan says so what? Impedance is high, and that’s a deal breaker. Localized heat creates a thermal problem, and power delivery could be challenging. There is, however, the advantage of glass being a better thermal coefficient of expansion match to the circuit board.
Really, in Swaminathan’s opinion, the only advantage to using glass interposers is the low cost IF you’re using large panel sizes. And then there’s the cost issue of establishing the infrastructure for manufacturing with glass. How long will it take before glass interposers can be viable?
For all these reasons, Swaminathan stands behind Silicon as a preferred substrate material. “We can’t just throw away silicon because it’s lossy. Some of the arguments being made for glass are that its less lossy, and lower cost, so lets just use it. I think that’s a wrong statement to make,” he said.
The real question is, where do you spend the development dollars? Swaminathan encourages the industry to spend the time and money reducing the cost of silicon, and developing the work arounds rather than getting the infrastructure in place to convert to glass because silicon is the better performer. Ultimately though, he says its the system-level architects who will drive one over the other. If they decide that glass is the way to go, the infrastructure will be created for it. ~ F.v.T
