Meeting industry needs with laser micromachining
10/01/2003
Overview
Laser micromachining is just beginning to be used in semiconductor manufacturing. Some of the more applicable areas are found in assembly and packaging. As the benefits become more apparent for this arena, laser processing will undoubtedly replace traditional mechanical and chemical techniques.
Lasers are finding applications for scribing, dicing, cutting, and drilling for a variety of assembly applications. In the past, the cost and complexity of incorporating a laser for micromachining has made this a tool of last resort. Today, the advent of compact, solid-state lasers with high output, particularly in the UV, has changed this choice. The laser is now gaining popularity in areas where mechanical machining methods are limiting device architecture, yield, or process throughput.
Laser processing vs. sawing
Laser processing delivers a number of key advantages over mechanical sawing and drilling, depending on the application. Two of the most important advantages are high yield and high throughput. For instance, edge and part damage from mechanical sawing always results in some scrap or defective product. In contrast, noncontact laser processing has been documented to deliver yields at 99.9% or better [1].
 Figure 1. Curves cut in silicon using a 355nm UV laser. |
null
The laser is also more flexible and versatile than mechanical tooling. A laser beam can be used to cut all types of patterns and shapes on demand, including complex curves (Fig. 1). Moreover, the contours of these cuts can be precisely controlled, even allowing, for example, reverse tapered via holes. Another major advantage of laser processing is that it does not require water or cooling fluids.
There are several areas within the industry where mechanical methods are barely adequate and will limit progress. These are the areas where the laser is now being investigated or is already accepted as the tool of choice.
Singulating copper low-k
The current trend to using more copper low-k interconnect structures creates potential die singulation problems. Conventional dicing saws can introduce large tensile and shear stresses at the cut zone that can result in significant cracking and chipping as well as adhesion loss between metal and low-k layers. Slowing the saw speed can mitigate some of these problems, but at the expense of production throughput. Moreover, some of the softer materials tend to stick to saw blades, thus clogging and slowing the process. It is also very important to consider that the porosity of low-k materials will absorb coolants used in cutting, so they have to be protected, adding to process complexity.
 Figure 2. The absorption spectrum of a typical silicon sample. |
null
An advantageous alternative is to use a laser to scribe completely through processed layers before using a saw to cut through the substrate to complete die singulation. This uses the mechanically benign laser process to create edges that act as crack stops. Two separate cutting geometries can be used. In the first, the laser scribe proceeds down the middle of each street, creating a trench that is wider than the subsequent saw cut. Alternatively, the laser can be used to create narrow scribes on either side of the intended saw cut. This second approach requires more individual laser scribes, but is potentially faster and more economical since it requires less material to be removed and hence lower laser power and less processing time.
Thin wafers
Thin silicon wafers represent another area where mechanical methods are reaching their limits. Currently, there are areas in the industry that dictate thinner silicon: stacked packages, high-power applications, and specific applications such as smart cards. Some IC manufacturers are now looking to thin wafers down to ≤200µm, compared to a typical 600–800µm.
Singulating these wafers with a diamond saw leads to a high degree of chipping and cracking because of mechanical vibration. One way to eliminate mechanical trauma is to singulate die by cutting through the entire wafer thickness with a UV laser. For traditional thicker wafers, this is a slow process that is often uneconomic. But as wafers get thinner, the laser has to remove less material, allowing faster processing speeds. At the same time, mechanical saw-yield problems increase as wafer thickness decreases. Depending on factors such as die value, there is a point at which the laser delivers higher net throughput and becomes the more economically attractive solution. The laser is now also being investigated as a tool to create microvias for stacking ICs from thin wafer.
Compound semiconductor substrates
With LEDs and integrated optical telecommunications components on compound semiconductor substrates, there are two issues related to cost and fragility. Specifically, fabrication complexity results in an extremely high cost for wafers after processing, even higher than the densest silicon circuits. But unlike silicon, many of these materials have hexagonal crystal structure resulting in a high loss due to fracturing and chipping by mechanical scribe and break singulation methods. The risk of chip and fracture damage also limits the ability to nest small die. Scribing with a UV laser, followed by mechanical fracture, avoids these limitations.
Interestingly, many compound semiconductors are fairly transparent at typical UV laser wavelengths (355nm and 266nm). However, even with a low absorption coefficient, the high peak power of the latest short-pulsed lasers means that a tightly focused beam is absorbed at these wavelengths due to nonlinear optical effects in a runaway process.
Choice of wavelengths
All solid-state lasers produce output in the near IR (l = 1064nm) that can then be converted to green (532nm), UV (355nm), or even deep-UV (266nm) inside the laser head. Today, most micromachining applications use either IR or UV lasers. In general, the shorter the wavelength of the laser, the higher the cost and the lower the power. So, from a cost/laser-W standpoint, it is always best to use the longest wavelength that will give the desired result.
Near-IR lasers deliver the highest raw power and hence the highest possible processing speed. For example, conventional thickness silicon wafers can be cut at speeds up 200mm/sec and scribed at up to 800mm/sec. However, at this wavelength, the material is removed by intense local heating (i.e., boiling). This thermal processing has two potential drawbacks: it produces a small amount of splattered and recast debris and can cause some peripheral damage to the substrate. As a result, near-IR lasers are generally used with difficult materials where edge quality is not of paramount importance.
 Figure 3. A high-aspect-ratio groove in a sapphire substrate made with a 266nm UV laser. (Courtesy JPSA Inc.) |
null
Many emerging applications use UV laser light at 335nm or 266nm. A typical example of this is scribing blue LEDs. There are two advantages when using UV wavelengths. First, the laser cuts by directly breaking atomic bonds rather than by heating, resulting in very little thermal damage to surrounding material. Second, shorter wavelengths can be focused to smaller spot sizes on the work surface. As a result, UV laser micromachining can produce small features with much better edge quality than longer wavelength lasers. The main disadvantage of UV lasers is lower power, and hence, reduced throughput.
Until recently, 355nm lasers were preferred over 266nm lasers, almost exclusively, because 266nm lasers offered limited output power and lower reliability. This situation has now changed with the advent of a new generation of high-reliability 266nm lasers.
Choosing the appropriate laser is also a function of the bandgap and the absorption-penetration properties of the material to be machined (Fig. 2): The penetration depth of near-IR laser light is around 104¥ greater than for a UV laser beam at 355nm. If the laser is to be used for drilling, scribing, or dicing applications, then this may not be a critical consideration. But if the laser is being used to create surface features, and it is important to avoid sub-surface damage, then a UV laser must be used.
You must also consider pulse duration. Peripheral thermal damage at any processing wavelength is minimized by using the shortest possible pulse duration, which reduces the heating cycle at the work surface. The best laser for micromachining combines a short pulse length (for best quality cuts) with a very high repetition rate and high output power (for maximum throughput). In response to this growing market need, laser manufacturers have designed lasers that precisely meet these performance requirements (e.g., the Spectra-Physics high-intensity and peak power oscillator, HIPPO, laser).
Dicing blue LEDs
Manufacturers of high brightness, blue LEDs are concentrating on higher throughput and lower production costs. The UV laser is having a significant impact in blue LED production [2].
 |
null
A blue LED consists of GaN epitaxial layers on either a sapphire or silicon carbide (SiC) substrate. Both of these substrate materials are difficult to cut and dice. Combined with small die size, this results in low yield and low throughput when using mechanical die separation. With conventional processing, these wafers are thinned to ~100µm, scribed with a diamond tool, and cleaved.
null
More recently, these die can be successfully diced using a 355nm laser; the laser is used to produce a narrow "V" groove in the wafer — to 30–50% of the wafer thickness — before cleaving with a fracturing machine. Not only does the laser reduce manufacturing costs, but it also results in a higher yield because the V-shaped groove concentrates the mechanical stress during fracturing, producing a much more controlled localized effect than mechanical scribing.
Until recently, this laser process resulted in a small (up to 10%) reduction in final LED output that was attributed to the fact that sapphire is fairly transparent at 355nm and some of the 355nm laser lights penetrate into the epi layer. (The sapphire is easily machinable at 355nm only because a tightly focused laser beam is intense enough to drive multiphoton absorption.)
In response to this situation, work at JPSA has pioneered the use of a new 266nm laser for this application (Fig. 3). Sapphire absorbs this wavelength more strongly at all power levels (Fig. 4), eliminating the penetration problem. Data gathered at JPSA show that this eliminates LED power loss due to sub-surface laser damage [1].
Today, when dicing blue LEDs at JPSA, the dicing operation cost is <$2 per 2-in. wafer, with uptimes exceeding 99%. Compared with mechanical scribing, wafer throughput is increased by 500% with a 97% reduction in overall operating costs. Together with a 10% increase in product yield and a 15% increase in die count/wafer, these numbers make an overwhelming argument for the laser process (see the table).
Solar cells, etc.
Silicon solar cell manufacturers are always striving to improve production and products in three areas: cost, efficiency, and longevity. These are all key to making solar electricity a more viable option. Recently, laser-based micromachining has played a key role in all three areas at Exitech [3].
Single-crystal silicon solar cells are typically produced on 5-in2 substrates fabricated from 6-in.-dia. crystals. These typically have a thin top layer of very hard silicon nitride. A laser is used to create a frontside groove (e.g., 20µm wide by 30µm deep). This groove is subsequently metallized to form electrodes with a higher cross-sectional area and better conductivity than conventional planar electrodes. The adherence of the metal deposit in the groove is important to the solar module assembly process and device longevity, since solar cells cycle between ambient temperature (-20°C in some locations) and 60°C in bright sunlight.
 Figure 5. Grooves in a solar cell following a surface etch to remove cutting debris. (Courtesy Exitech Inc.) |
null
The IR laser is ideal for this application. Its high-pulsed power cuts easily through the SiN and silicon, and the high average power (10s of W) and pulse repetition rate (~100kHz) enable this application to be economically executed.
For this particular process, throughput and speed are everything and high edge quality, which could be achieved with a UV laser, is not needed. Typically, the required grooves are cut at speeds of up to 1m/sec in crystalline silicon. The total groove length is around 14m/wafer, which translates into a processing time of just a few seconds/.wafer. Moreover, the latest high-repetition-rate lasers enable 80% pulse-to-pulse overlap, which produces smooth, clean grooves (Fig. 5).
The future of lasers in assembly
Laser micromachining for semiconductor applications is just beginning to be used by manufacturers. At the end of 2002, there were several hundred lasers installed worldwide for this use. However, as the benefits of this technology become more apparent, laser processing will undoubtedly replace mechanical and chemical techniques in a wider variety of assembly and other applications.
Andrew Held, Mingwei Li, Spectra-Physics, Mountain View, California.
References
- J. Park, P. Sercel, "High-speed UV Laser Scribing Boosts Blue LED Industry," Compound Semiconductors, December 2002.
- Conversations with Jeff Sercel, president, JPSA Inc., Hollis, NH, a laser integrator that provides turnkey laser workstations to the semiconductor industry and contract laser-manufacturing services.
- Conversations with Herbert Pummer, president, Exitech, Foster City, CA, a turnkey system integrator and supplier of job shop services to the electronics and semiconductor industries.
Andrew Held received his PhD from the U. of Pittsburgh. He is director of marketing at Spectra-Physics, 1335 Terra Bella Ave., Mountain View, CA 94043; ph 650/961-2500, fax 650/964-3584, [email protected].
Mingwei Li received his PhD from Ohio State U. He is engineering manager at Spectra-Physics.