Amkor knocks down back-end operation costs
11/01/2001
Amkor knocks down back-end operation costs
With a $50 million investment in a program that began in 1999, Amkor Technology, Chandler, AZ, has now fully implemented a high-density semiconductor back-end process that integrates package, assembly, and test.
 Amkor's newly implemented matrix assembly and test process has moved the "wall" separating conventional assembly and test, improving productivity. |
Scott Voss, VP of Amkor's leadframe business group, tells Solid State Technology, "The benefit here is a reduction in the cost of materials, labor, cycle time, and floor space, and a significant increase in productivity. In addition, this is a dramatic change in the way most popular IC packages are assembled and tested."
He adds, "Although we have begun this program with leadframe-based packages, we are also working on laminate and tape-based chip-scale packages. There is no reason why higher pin count packages can't use the same concept to reduce assembly and test costs."
A visit to Amkor's assembly lines would reveal several key differences in the conventional concept of IC assembly and test work flow (Fig. 1); the new Amkor matrix assembly and test process uses:
- a common 70mm x 250mm leadframe-strip format (dubbed HDLF or high density leadframe);
- interchangeable automold chases that provide flexible capacity;
- parallel test in a strip format before package singulation;
- a common strip test handler with "quick change" contactor boards, and
- high pin-count testers that enable more parallel testing.
With the enlargement and standardization of copper leadframe strips to 70mm x 250mm, Amkor can assemble and test up to 585 IC packages on a single leadframe, depending on package size. For example, the industry's conventional leadframe format would only accommodate five 20mm x 20mm 144-lead LQFPs, but the new HDLF format accommodates 16. In another example, for a 16-lead TSSOP package, the number of units/leadframe strip increases from 50 to 168. This process is applicable to most leadframe-based IC packages, which account for 90% of ICs used today.
Integrating electrical test into the assembly process while IC packages are still held in strip format allows up to 48 eight-lead SOICs to be contacted simultaneously and electrically tested in parallel on a 512-pin tester, reducing indexing and test time/unit.
Voss explains, "This high density process is the packaging equivalent of producing a 300mm silicon wafer in terms of throughput, and equipment and facility utilization. We are able to process more packages in less time using less factory space and with less handling."
For example, in conventional assembly, a $1 million automolding system set up for 16-lead TSSOPs would produce 600,000 units/week using the old leadframe format. By contrast, it can now produce 2 million units/week using the new HDLF format with 168 units/leadframe strip. Thus, the cost/unit of capacity is 1/3 more than the old manufacturing format.
Amkor engineers have seen a 6x increase in throughput from trim-and-form machines, a 30-40% reduction in molding compound use, and typically a 20-25% reduction, depending on complexity, in costs associated with copper leadframes.
Operational advantages mean that automatic molding equipment can be converted from one package size to another in >1hr. "In the past, automold systems were typically dedicated to a specific package body size and were never changed to another package. Thus, if there was no production to run, the automold just sat there," says Voss. The process has also eliminated the need for test handler change kits. "With the right tester, one strip test handler can replace up to 10 singulated handlers, each of which uses change kits."
With the new process, test operations take about half the floor space for a throughput previously achieved with conventional processing. Joe Holt, VP of Amkor's test business unit, says, "Most ICs today are still tested in singulated format, using a dedicated test handler. With the strip testing that we use in our new process, a large number of devices are tested in parallel, lowering test cost as much as 80% depending on test time and the number of units contacted at one time. We also see increases in throughput up to 10x or more and improved quality since leads are formed after test instead of before. Bent leads are virtually eliminated, and it is much easier to contact all the leads on a package while they are still flat as opposed to trying to contact formed leads."
Amkor has converted more than 50% of its manufacturing capacity to matrix assembly and test; it is now in production at facilities in Korea and the Philippines and will soon be introduced in Japan and China. The process has been applied to more than 25 package types from eight-lead SOICs to 176-lead LQFPs.
Elusive nanopattern formation seen "in the buff"
Researchers at the US Department of Energy's Sandia National Laboratories have observed — seemingly for the first time — deposited atoms forming orderly, controllable two-dimensional nanopatterns.
 An evolution from droplets to stripes to inverted droplets, predicted by theory, in successive images by Sandia researchers. |
Scientists at the Albuquerque, NM-based labs think that such an assembly process along with control over physical factors influencing the process (e.g., temperature and material composition) means that nanotemplates can be formed to fine-tune device characteristics of self-assembling nanostructures. One possible end application could be tailoring photonic-lattice devices for controlling light in telecommunications. A low-energy electron microscope, showing exactly how nanostructures are generated, self-assembled, and transformed, showed atoms of lead deposited on a copper substrate, forming lead dots, then lead stripes, and then reverse dots (i.e., copper becoming the dot material) as more lead was added (see figure).
The work at Sandia, described in a recent issue of Nature, has produced a video recording of atoms self-arranging themselves in a manner long predicted by theorists. Sandia physicist Norm Bartelt says, "Such theories generally had been treated with a great deal of skepticism. There was no obvious route for atoms to arrange themselves in predicted patterns." Simply described, kinetics predicts that 10,000 moving atoms should go anywhere, without an assembly arising. But theorists have long believed competing attractive and repulsive inter-atomic interactions can lead to the spontaneous formation of ordered patterns in widely varying chemical and physical systems.
Project leader Gary Kellogg notes, "This work, which to our knowledge is the first unambiguous observation of the expected sequence of domain patterns, helps understand the new physics that manifests itself at these small length scales. New materials with highly specialized properties necessary to meet defense and consumer applications can be fabricated only by tailoring the structure of materials on the nanometer scale. This work provides insight into how nature does this, and how humans can do the same. The close agreement between experiment and theory allows us to probe the key inter-atomic force parameters involved in the process."
Sputtered films: Targeting SiN with pure Si and an S-Gun
Researchers at Sputtered Films Inc. (SFI) believe they have developed a reactive sputtering SiN process that simultaneously eliminates irradiation damage occurring at p-n junctions while creating a robust hermetic seal that prevents moisture damage to previously deposited film layers.
The resulting SiN process will be particularly important in PVD encapsulating applications such as light-emitting diodes (LEDs). This is because p-n junctions, being very close to the wafer's surface and very thin, are extremely sensitive to irradiation. Irradiation causes an increase in device forward voltage or even total device failure, according to SFI president and inventor of the S-Gun Peter Clarke, engineering manager Pavel Laptev, and senior scientist Valery Felmetsger.
They attribute the process results to the ability of the company's new DC-RF-coupled S-Gun sputter source to simultaneously control the cathode power, RF power, heater power, gas flow, and gas composition, as well as the use of a pure Si target in lieu of an SiN target.
"By using a DC-RF-coupled S-Gun-type magnetron, we have been able to protect the target from insulating film growth [target poisoning] and so eliminate arcs from the target — a major cause of structural defects on the substrate," comments Laptev. "The user can also fine-tune the process result, unlike the case when conventional magnetrons are used."
"When you use a pure Si target, you can adjust the nitrogen flow to get a stoichiometric film and reach a high deposition rate," states Clarke. "You can't do that with an SiN target without incurring unacceptable productivity problems. For example, the deposition rate is about 100 times lower than what is achieved using a pure Si target."
The operating paradigm of the S-Gun magnetron is key to the results achieved. "There is no ion bombardment of the growth film so irradiation damage is eliminated," notes Felmetsger. "The magnetic field structure of the S-Gun magnetron confines DC and RF plasma in the target vicinity. As a result, there are no field lines going to the substrate to provide charged particles." According to Felmetsger, this last benefit is critical because if an LED device experiences ion bombardment, it can have its forward voltage increased beyond the acceptable limits of 0.2-0.3V. "SFI's process increases the forward voltage by no more than 50mV."
What's more, SFI reports that its PVD silicon nitride process has a very low hydrogen level, unlike the PECVD process, which can produce films with up to 40% hydrogen content. The hydrogen content of a film is critical, notes Felmetsger, because high levels lead to film structure degradation in subsequent thermal processing steps. "There are a number of vacuum deposition methods that are free from hydrogen," says Felmetsger. "But RF sputtering's deposition rate is too low for manufacturing applications, AC sputtering, while having a high dep rate, results in quite a bit of irradiation damage."
Another advantage the company sees in using its S-Gun is the specially designed anode. "The reactive process creates a dielectric layer on the anode, which is the cause of the phenomenon known as the disappearing anode," explains Felmetsger. "Using a multi-disk configured anode ensures permanent conductivity for the magnetron discharge current and prevents arcing on the anode."
Less power to the OLEDs
Electrically efficient organic light-emitting diodes (OLEDs) are being explored as an energy-efficient alternative to fluorescent and incandescent bulbs for lighting buildings and homes. Researchers from the US Department of Energy's Los Alamos National Laboratory, headed by Darryl Smith, are developing ideas to increase power efficiency in OLEDs. At the 222nd national meeting of the American Chemical Society in Chicago, Smith addressed the use of a chemical layer, a molecule thick, to facilitate the flow of current from a power source into self-assembled polymers to increase efficiency.
OLEDs are used in flat display screens, such as vehicle dashboard readouts and in postage-stamp-sized data screens built into pilots' helmet visors. Arrays of OLEDs just one millimeter thick can display full-color moving or stationary pictures and graphics. Because OLEDs emit their own light and can be incorporated into arrays on thin, flexible materials, they also could be used to fashion large, thin panels for light sources in buildings.
While OLEDs show promise in a number of applications, their biggest drawback has been the power required. Efficiency is the amount of light in lumens, integrated over the spectrum of eye response, per power in lumens/watt. For current display applications, 30 l/w for OLEDs is sufficient. In fact, this figure is better than incandescent bulbs (15 l/w), approaches fluorescent lighting (80 l/w), and is comparable to inorganic LEDs (40 l/w). OLED efficiency is within the ballpark of being interesting, noted Smith. If you can make OLEDs even more energy efficient, then you could potentially create very bright, very long-lasting light sources that would be versatile, flexible and inexpensive to use.
A standard OLED consists of a transparent layer of electrically conducting material, such as indium tin oxide, deposited on a transparent substrate. On top of this first conducting layer is a layer of organic polymer — a chain of carbon-based molecules — that emits light when excited by electrical current. A second conducting layer is deposited on top of the polymer.
When voltage is applied across the conducting layers, a current runs through the polymer layer, which emits photons, creating a spot of light. Because of the different chemical and physical properties of the conducting and polymer materials, it is difficult to create a smooth flow of current through the layers. As a result, more voltage must be applied to achieve the current required to generate photons from the organic polymer.
Smith and Los Alamos colleague Ian Campbell have shown that an intermediate chemical layer can be applied between a conducting layer and the polymer to achieve more efficient current flow. The process uses what nanotechnologists call a self-assembling monolayer to facilitate electrical flow between OLED layers.
The intermediate self-assembling monolayer consists of rows of molecules that line up in the same direction when applied to a substrate. By adding atoms at both ends of a monolayer molecule, the researchers developed a molecule that would anchor itself to the conducting layer while maintaining a distinct polarity across the entire molecule, in essence charging the ends of the molecule like the poles of a bar magnet. The result is a molecular layer a few billionth of a meter thick that helps shuttle electrical charge between an OLED conducting layer and the polymer layer.
In order for OLEDs to be used as light sources, their efficiency must exceed existing fluorescent light bulbs. That means current OLED efficiency must triple. Still, the gains in efficiency shown by use of a monolayer in OLED manufacturing has illuminated interest in the idea and the feasibility of decorative panels of light that can be operated inexpensively by the consumer. With recent concerns over energy supplies and costs, the research is particularly relevant.
OLEDs for displays are already becoming commercially available. True lighting applications are years away. In five years, prototypes are likely, according to Smith, but justifying the cost of replacing inexpensive conventional lighting will remain a challenge.
Dropping ECD's heavy baggage
NuTool engineers have shown that overburden associated with conventional electrochemical deposition (ECD) can be reduced from 500-2000nm to 100-500nm using the company's electrochemical mechanical deposition (ECMD).
As a result of the significant decrease in unneeded copper deposition, the company expects the associated COO (cost-of-ownership) of CMP to be reduced $7-$15 from its typical range of $12-$35. In addition to having less material to remove, the savings would come from improved yield due to reduced erosion and dishing.
"The ECMD process provides a greater level of planarity before the CMP process even begins," states Brian Stickney, NuTool director of sales and marketing. "It achieves global planarity and local planarity. In many cases, the user of the ECMD process starts out with better planarity before CMP than could be obtained after CMP if conventional ECD had been used. Customers typically see equal or superior electrical performance of the die with a reduction in CMP COO that ranges from 40-60% before process optimization. At a minimum, ECMD provides equal results to ECD copper deposition with a significant cost reduction for copper CMP."
Because dishing and erosion add variation to the electrical performance of a device, achieving a highly planar film enables uniformity of performance. Interestingly enough, NuTool believes its ECMD process could be a market buster for CMP. "If the overburden is low enough, a wet process can be used instead of CMP," comments Stickney.