A Rose by Any Other Name . . .

Twenty years ago, as a young packaging engineer at Western Electric's Engineering Research Center, I was startled out of my office by a commotion down the hall. A group of 10 engineers and managers was gathered around a conference table, with Jack Balde at the center, showing off a mockup and expounding on what he thought would be the "next great thing."

When the chief architect of AT&T's introduction of modular telephone connectors and flat, under-carpet wiring — and the man who championed the use of chip carriers and surface-mount technology within Western Electric — spoke, the rest of the company listened.

The next great thing, according to Jack? Multichip modules, or MCMs. The idea was to increase the functional density of circuit boards by combining related function chips on a common high-density substrate. It didn't matter what the specific substrate was. What mattered were small islands of higher density that would result in systems with higher performance and lower cost. Jack's vision compelled me, and many others like me, to become "MCM technology pushers."

Twenty years later, MCMs are still seen as the next great thing, although the names have been changed to protect the innocent. Now we hear MCP (multichip package), FCP (few chip package), SoP (system on a package) and SiP (system in a package). Maybe things are different now.

Two-chip stacked MCM with logic and memory in a BGA footprint. (Source: ChipPAC Inc.)

In the early 1980s, IBM introduced its TCM (thermal conduction module) with 121 ECL chips on a 32-layer HTCC substrate. The huge cost of the infrastructure and components was a small price to pay for increased performance when a three-board CPU was shrunk to one board.

In the early 1990s, Hughes, Boeing and other aerospace suppliers introduced complex 10-30 chip MCMs on 2 × 4 in. thin-film polyimide multilayer substrates for radar and communications. The savings in size and weight for these products, which flew on planes, missiles and spacecraft, easily justified the higher cost of these modules.

But widescale use of MCMs was held back because there wasn't a compelling need to justify the higher perceived risk and cost of an MCM solution. When the performance gain is less than a factor of two, it is a harder sell to risk your product line on a new technology with a weak infrastructure.

Many startups and spin-offs bit the dust by betting on MCMs being the next great thing. Polyicon, Advanced Packaging Systems, nChip, MMS, Alcoa's AVP and the GE-TI joint venture are just a few.

Maybe things are different now. For the first time, I've been hearing the words "faster time to market" and "cheaper" to describe MCMs, not by the MCM technology pushers, but by the end users.

Every cell phone manufactured today has an MCM — a two-chip stacked module with SRAM and flash. Every notebook computer manufactured today with a graphics controller has an MCM. It's a graphics chip with an SDRAM. The performance driver in these cases is a smaller footprint for a shrinking form factor consumer product. Of course, the implementation is with the lowest-cost substrate and assembly technology.

IC technology will always increase in integration levels. Last year's design that required four ASICs will be implemented in one ASIC this year. SoC (system on a chip) is just the name of the next generation of chip integration, which uses larger functional building blocks in the library.

But the one application where an SoC will never compete with an MCM solution is integrating mixed-technology chips. When a logic chip typically uses one poly layer and five metal layers, and a memory chip typically uses two metal layers and five poly layers, the MCM integration will always have a cost advantage.

The secured niche for high-volume MCM applications is for the combination of a large logic chip, such as an ASIC, microcontroller, microprocessor, graphics chip or DSP engine, with one or more memory chips, such as SRAM, SDRAM or flash. The supplier of this MCM gets to sell more silicon, and the user gets a smaller form factor, slightly higher speed, lower cost than any alternative and a single-chip package they are used to dealing with.

MCMs are finally becoming market-pulled. Maybe things are different now.