Dicing takes a finished wafer (post-wafer fabrication) and converts it into individual dies; it’s the step where the front-end fab process transitions to the back-end assembly process. Grinding & Dicing Services, Inc. has accumulated 25 years of engineering and process knowledge supporting semiconductor, consumer electronics, and medical companies worldwide. We have identified the five most common obstacles getting in the way of a successful transition from the prototype to production of a MEMS part:
- Residual Stress
- Cost Per Die
- Size Reduction
Keeping It Clean
MEMS devices are designed to interact with the environment, interpreting the analog world. They are particularly susceptible to contamination, water and air pressure, physical shock, and vibration. Blades and ablation lasers dig up the material in the streets creating particles. Water or air is used to keep the particles suspended and washed away.
As an added measure, wafers are often coated before MEMS dicing. Dicing tape is applied on the backside of a wafer when blade dicing (“mechanical dicing”) is the chosen method. Stealth dicing has become the new standard for safe release of sensitive MEMS and sensor design, primarily due to its dry, subsurface laser process which does not create particles or liquid contact with the devices. The dies separate during an expanding step, where care is taken not to create slivers of silicon.
Comparatively, plasma dicing is a somewhat new method of applying the Bosch etch process. When done correctly, the dies are particle- and contamination-free. This requires a custom mask design for effective plasma dicing, which often means that alternative dicing methods will be rendered unfit as a backup strategy.
Plasma is a process which holds a lot of promise; a novel front-end technique but one not easily integrated into backend processes. We often limit the chemicals allowed in the backend to DI water and IPA rinse for good reasons. Introduction of C4F8, SF6, O2, Ar, N2 gases for plasma dicing is in many cases impractical for the backend.
Do No Harm
Wafer dicing, cleaning, expanding, and picking look very gentle at the macro level. At the micro level, the level at which MEMS interact with the environment, those same processes can be quite violent with the die. Protecting the die (encapsulation of the layer using a carrier wafer or some form of protective polymer layer) or choosing a low impact process are the obvious choices.
Most mechanical dicing processes have multiple negative impacts to consider. Blade dicing vibrates the devices with both the cutting motion and water contact. Coolant water is under high pressure and can destroy delicate MEMS. Optimizing the cutting process and choosing the right blade greatly minimize the vibration from the cut. Damage done by the water is much more difficult to control.
Covering the wafer with tape can reduce the damage from water pressure. Protective tape can also limit the damage caused by the vibration from water nozzles meant to cool the blade and remove particles. Flooding instead of spraying with water can help with some of the more robust devices.
Laser ablation dicing is non-contact and therefore gentler than mechanical dicing. Ablation begins at the top surface and removes material layer by layer. “Cold ablation” is a popular term which refers to laser pulses being shorter than the thermal conductivity of the material being ablated. Heat is not transferred to adjoining atoms leaving the spot “cold.” The silicon being ejected is thousands of degrees which can heat the device creating a Heat-Affected Zone (HAZ).
This super-thermal process ejects material at supersonic speeds and creates shock waves which can damage fragile layers and devices.
Thinning the device reduces the number of required laser passes, and thus reducing the damage. Enlarging the device moves the laser away from sensitive areas allowing the heat and shock waves to dissipate.
Stealth dicing avoids shock done to the active surface and redirects it underneath the surface, into the material. A line of pulses (e.g., modification layer or “SD layer”) creates a controlled crack plane along which the wafer eventually separates into dies during the expansion step. The closer the line is to the surface, the lesser the impulse energy needed to separate the die. The farther below the surface a line is, the farther away the laser shock wave is from the active area. Being that there is no material ejected and so little heat generated, there is no heat transferred to the device.
Plasma dicing holds a great deal of promise as a non-contact, low-temperature, and low-stress process. Etching is a blanket process which requires masking for areas not meant to be etched. Silicon in the dicing streets needs to be exposed, and non-etchable materials such as metal need to be removed from the streets. Blade or laser scribes are effective options for exposing the silicon and removal of the non-etchable materials, thereby creating the mask. Unfortunately, use of blade dicing or laser ablation dicing negates much of the plasma dicing advantage.
For plasma dicing to fit into the standard process flow, it needs to be done on dicing tape and frame. Great care must be taken to ensure there is no air bubble between the wafer and tape. With blade, ablation, or Stealth dicing, a bubble in the dicing tape can pose minor risks to the wafer. However, for plasma dicing, any bubble between the dicing tape or masking film and the wafer can lead to arcing. In the worst-case scenario, plasma discharge arcs will burn a hole through the wafer. Non-contact vacuum tape mounting, visual inspection of each wafer for bubbles, and a temperature sensor on the plasma chamber chuck table to detect hot spots can reduce the risk of burn holes in the wafer.
Residual stress can affect the performance of certain types of MEMS. Limiting top and backside chip-out from blade dicing can reduce residual stress in the die. Thermal stress from ablation dicing can be reduced by thinning the wafer (fewer required passes), moving the dicing street away from the device, and allowing for an angled sidewall.
When applied properly, both Stealth and plasma dicing leave no residual stress. Stealth places stress below the surface in the dicing street, away from the active die. During expansion, the stress is released, yielding dies with no measurable residual stress. Likewise, plasma dicing can leave no residual stress. “Undercutting” is a common quality issue to watch for with plasma. It can leave the die stressed along with other problems.
Cost Per Die
When dicing MEMS wafers, there are five variables that dictate the cost per die: number of die per wafer, time to dice the wafer, initial cost of the system, consumables, and yield. Narrowing the dicing street increases the number of die per wafer. Narrowing the scribe lane is not always possible due to the structures in the street itself. When the streets can be narrowed, Stealth dicing and plasma dicing allow for streets less than 20µm wide. It is difficult to reduce the street width for blade dicing or laser ablation for a host of technical reasons. Blade dicing is limited by the edge chipping and blade wear. Laser ablation requires a wide street to eject material unless the wafers are thin.
Clearly, the less time it takes to process a wafer/die, the lower the cost per die. For most MEMS wafers, the fastest way to dice is using Stealth. When passivation is designed in as a mask, and the dicing streets are clear of metals, plasma dicing can win out. If a mask needs to be applied and metals removed from the street, blades or a laser are applied to complete the trenching. Consequently, the overall plasma dicing flow is a very long process at which point residual stress can be introduced into the material. Blade dicing for wafers with tens of thousands of die takes a large amount of time simply because of the large distance the blade needs to travel.
By far, the lowest initial and ongoing cost option is blade dicing. This has been the semiconductor industry standard for decades. Readily available used systems and spare parts are ubiquitous. Stealth dicing is the next most cost-effective strategy, followed by laser ablation.
The up-front capital outlay for the plasma solution represents the biggest hurdle of all options while the average cost per wafer can vary relative to its peers depending on the application itself. Ongoing costs for plasma are high though made manageable for high-volume programs (inflexible for R&D and impractical for low-volume runs).
The expected yield based upon each singulation process is a difficult comparative model to use because there are so many different types of MEMS. In general, plasma dicing offers excellent yields due to its precision and overall etch quality. No process is perfect; even with plasma. As discussed earlier, a bubble between the wafer and tape can lead to an arc in the chamber burning a hole in a wafer. A bad reaction to the masking process can greatly reduce the yield.
Stealth dicing is a close second for high yield. There are some devices that are susceptible to damage from scattered laser light that require dicing from the top side or changing the process depth. Ablation dicing can give good yields when the beam is positioned far enough away from the device and the surface of the die has a good protective coating. Blade dicing tends to have the lowest yield. It will depend upon how sensitive the die is to particles and vibration. For robust devices, blade dicing results in high yields.
There is a constant push for smaller and smaller die with greater functionality. For dicing, as the die shrink, the number of die on a wafer increases. It isn’t uncommon for the number of die on a wafer to nearly double over time. Die size is set by many factors: the size of the active area, edge space required for bond pads, street width for test pads, and dicing. Designers use the term “core limited” when the active area of a die sets the die size. The term “pad limited” is often used when the area required to accommodate the number and size of pads set the die size. Dicing street width cannot go below what is needed for the test pads and dicing tolerances.
Blade dicing has the most challenges when shrinking the die or street. Doubling the number of die doubles the time it takes to dice the wafer. That doubles the amount of time the dies are exposed to the corrosive water and vibration.
Being a contact dicing method, the smaller the die, the more likely they are to fly away. Because street width must consider the amount of chipping, dicing streets will always be wider than with other methods, leading to larger die. Blade life may also be a factor. In some cases, the reduced blade width combined with the increased number of dicing streets results in blades not lasting a whole wafer.
Laser ablation has the challenge of removing material from the kerf. A 20µm-wide kerf on a 500µm-thick wafer is the same ratio as 14 inches between two three-story houses. Try and picture throwing a ball out of that small gap without hitting either one of the houses. It is very difficult. Some of the newer ultra-short pulsed lasers can do this but are cost-prohibitive.
Stealth dicing doesn’t care what the die size or street width is, producing a die of the smallest dimensions. When dicing from the backside, you can choose to dice in the middle of the street or to one side. This allows Stealth to dice wafers designed for blade dicing or Stealth dicing. As the die get very small, tape expansion may not have enough energy to separate the die. In that case, a breaker like the ones used for LEDs is then added to the process flow.
Plasma dicing likewise does not care what the die size or street width is. What is required is a mask applied or designed from the front end with the passivation and the metals removed from the street with a blade saw or laser.
Contact us today to learn more about our Stealth dicing services. We are based in San Jose, CA, and work with clients worldwide.