The inexorable march towards smaller, faster, and more capable electronic systems has been breathtaking. In 1946, ENIAC, the first programmable computer, housed 50,000 vacuum tubes in 80 feet of cabinetry, drew 150 kilowatts of power, and performed 5,000 operations per second. Today's off-the-shelf Pentium microprocessor jams 2 billion transistors onto a 2.2 square centimeter sliver of silicon roughly 0.3-to-O.7-millimeter thick, draws 80 watts of power, and can perform 3.2 billion operations per second.
The development of the integrated circuit, independently invented by Jack Kilby of Texas Instruments and Robert Noyce of Fairchild Semiconductor, in the late 1950s placed the information industry on this torrid pace of innovation. For the past 35 years, the number of transistors on integrated circuits has doubled approximately every two years. This doubling effect is the so-called Moore's Law, named after Intel cofounder Gordon Moore.
These thousands, then millions, and now billions of transistors switching on and off generate heat. In today's most advanced systems, silicon chips--called dies by their manufacturers--operating at 85 [degrees]C can generate average heat fluxes of more than 100 W/cm2 and produce localized, submillimeter hot spots often exceeding 1 kW/[cm.sup.2]. This is within an order of magnitude of the heat released into space by the surface of the sun. Without our ability to remove ever-greater heat fluxes from the surfaces of integrated circuits and other electronic components, we would never realize the benefits of their prodigious computational capability.
Engineers often specify air-cooled heat sinks or liquid-cooled cold plates to stabilize high-flux chips thermally. They are attached by successive layers of heat spreaders (usually copper plates that diffuse heat over a greater area) and thermal interface materials (often thermally conductive particle-filled silicones or greases).
As the performance of microprocessors has begun to approach the complexity of the human brain, the three-dimensional architecture of nature's most powerful biological computer has inspired new ways to organize dies. One promising approach is the three-dimensional chip stack. Here, adjacent chips are piled directly above one another, typically separated by 10 to 50 micrometers, rather than located next to one another and separated laterally by 10 to 50 millimeters on a printed circuit board.
There are several reasons why chip stacks will help us maintain the cadence of Moore's Law far into the future. The vertical placement of one chip on top of the other provides a third dimension of interconnected transistors and functional electronic macrocells. Such close proximity--micrometers rather than millimeters-nearly eliminates significant time delays as signals travel between chips.
Equally important, fusing chips with different functional capabilities--processing, memory, power, communications, and environmental sensing--into a single chip stack could lead to compact microsystems of unrivaled capability and truly ubiquitous computing.
Engineers have already begun to design products to leverage those advantages. Stacks of memory chips have been in commercial use for more than ten years. Yet rudimentary stacks of two or three low-power logic dies are...
This is a preview. Get the full text through your school or public library.