June 25, 2018
Interposer made at Georgia Tech
The penetration of glass into mass production has been slow so far, but could the pressure to obtain a higher flow rate be lower if the material plays a greater role?
For more than half a century, high-purity silicon has dominated the world of integrated circuit production. It’s a material that has almost everything.
Silicon is not only an efficient semiconductor, it is mechanically strong. This is in direct contrast with many III-V materials that offer high performance but are much more fragile.
Sapphire offers strength and excellent electrical insulation but is difficult to economically integrate into the manufacturing. As a result, materials such as gallium nitride, which are useful in light emitting diodes with high brightness and power devices, have passed from sapphires to silicon wafers.
A challenger emerges, but even based on silicon. The optimization of chemistry and manufacturing processes makes glass a serious competitor.
One advantage of glass is that it works well as a complement to silicon, allowing it to optimize manufacturing processes and experiment with new ways of eating in adjacent markets.
According to research firm Yole Développement, the use of glass as a substrate for microfluidics is the most widespread and will probably remain so in the early 2020s with the evolution of other applications of the material.
Although polymers such as polymethyl methacrylate (PMMA) have become popular for lab-on-a-chip devices because they are less expensive to process, they can not be used in many applications. Glass and pure silicon react much less.
The other important use of glass today is largely hidden. As integrated circuits were embedded in smart cards, hardware manufacturers needed to find ways to keep silicon wafers stable and flat while placing them down and then connect them to the top. Silicon lightens slightly and collapses to a thickness of only 100 μm. The solution is to temporarily bind the diluted wafer to a glass slide.
A major advantage of using glass as a support and substrate is that slight changes in the chemical composition allow the thermal expansion coefficient (CTE) to be combined with everything it needs to join. This, in turn, may cause the factory glass to pass through to the integrated circuit packages themselves.
At the beginning of the decade, the Georgia Tech Packaging Research Center began using glass in the 2.5D and 3D multi-chip modules of high-density smartphones, high-end network devices, and computers.
For 2.5D and 3D packages, manufacturers faced two different options. The first was to adapt existing organic polymeric materials used in low-cost IC packages to work as mini-PCBs interconnecting the multiple chips of the module. However, organic matter has two problems. Although they are inexpensive to use, the connection may be limited. Organic packaging tends to deform and stretch on heating and cooling, which can cause very thin joints to bend and break completely. As a result, there are limits to the fineness and proximity of the connections, which makes it difficult to design modules requiring very large buses between the peripherals of the package.
The other option that chipmakers have hitherto used on a commercial scale is the use of silicon as an interposer. Foundries such as TSMC used older production lines, typically those capable of producing 90 nm or 65 nm generation integrated circuits, for interleaves with a connection density similar to that of the chips themselves, but at a cost higher. The spacers are generally completely passive, although it is possible to place switches and other active devices on the substrate.
Georgia Tech’s PRC has proposed glass as a third option that can compete with silicon interconnect density and superior ETC compatibility, but with a cost structure closer to that of organic packaging. Unlike silicon, glass should not be manufactured as a wafer. An experience in the manufacture of LCD has helped develop a mature market for devices handling large rectangular panels. This allows more devices to be manipulated simultaneously without having to cut square devices in a circular blank. Glass suppliers have shown that they are capable of forming substrates 1 meter long, although 500 mm may become a common size for packaging operations.
Another advantage of the glass is that it remains mechanically strong, even diluted like silicon. Roll processing is possible thanks to the increased flexibility of the glass in extreme dilutions. The researchers worked on inserts as thin as 50 μm to support low profile housings.
Below: a schematic representation of the typical process of bonding and transfer of platelets for the thinning of silicon wafers
The scramble a problem
But they discovered that it was not just good news for the glass. The legendary fragility of glass and its low thermal conductivity could be problematic. Fleas are often designed with the idea that much of the heat they produce will escape into the PCB. Designers should rethink that if their devices were linked to a glass substrate.
Technology designed to increase the density of connectivity in 3D enclosures can help Glass to become the interposer of choice in helping to solve its problems. Silicon vias (TSVs) have built dense DRAM batteries that offer much higher throughput than conventional memory modules. The method includes drilling or etching holes in the substrate and filling the cavity with copper or a similar conductor to connect traces from the top to the bottom of each device in the stack.
Vias on organic laminates can not be spaced much closer than 200 μm. The TSV can be as close to each other as 10 microns. m, although most designs use more casual spacing, because relatively thick copper traces can easily penetrate the chip space that must be used for active logic. The work of the CPM and suppliers such as Corning have produced test equipment on glass with a 30 μm through hole.
Copper conductors could help both brittleness and thermal conductivity. Metal vias will help transfer heat more efficiently to the PCB. R & D teams suggested using additional copper structures. In dense piles, copper may release heat for spreaders on the edges of the package. Corning also found that thinned glass is less likely to break in areas containing vias.
Although high density glass can lead to the removal of organic and silicon interleaves in computer products, other properties of glass indicate the first applications of wireless communication. Glass is already used to make passive RF filters and similar devices and has advantages in RF connections. The Taiwan Industrial Technology Research Institute (ITRI) has tested interconnections suitable for RF operations up to 20 GHz. The group found lower losses in glass-based waveguides and microstrip lines compared to silicon substrates.
Its high resistance favors the use of glass as a support for integrated passive devices operating at the frequencies required for millimeter wave and 5G radios.
Several years ago, STMicroelectronics set up lines for the production of integrated passivation glasses. Corning and Qualcomm have been working on the construction of high Q inductances on glass using metal-insulator-metal architectures. The ability to pack many passive components under a SoC can also help to use glass for high density IoT controllers.
The intrusion of glass into mass production has been slow and almost invisible. But the pressure to get a higher throughput could make this material a much bigger player over the next decade.