Inside the Back End of Power Semiconductor Manufacturing: A Tour of Shiima Electronics

Open on YouTube ↗
Overview

After touring the front-end process at Toshiba Device & Storage's Kaga plant, where the chips inside power semiconductor packages are made, the YouTuber Ichiken visited Shiima Electronics in Yamanashi, Japan, to see the other half of the story: the back-end process. This is where finished chips on a wafer are turned into the packaged parts people recognize from real products. Shiima Electronics mainly does evaluation work on semiconductor prototypes, and it gave the channel a special opportunity to film how it handles the back-end steps for power semiconductors.

13 min read

The tour matters because of what kind of production it shows. The work at Shiima is small-quantity prototyping, not mass production, so many steps are done by hand. In a mass-production line, much of this happens inside equipment that works like a black box. Here, the video explains, you can actually see what is being done at each stage.

The Starting Point and the Sequence of Steps

The process begins with wafers that have already been through the front end. The devices are already formed on them and are, in that sense, finished. The back end then processes the wafer through a series of steps: backgrinding, dicing, die attach, wire bonding, resin encapsulation, and singulation. The video follows these steps roughly in order, and adds several supporting steps along the way, including tape changes, flux cleaning, plasma cleaning, laser marking, and inspection.

2:06

Backgrinding: Thinning the Wafer

The back end starts by grinding down the back side of the wafer, a step called backgrinding. First, backgrinding tape is applied to the front of the wafer with a dedicated machine. The tape holds the wafer in place by suction and protects the circuits on the front from debris during grinding. To keep the tape firmly attached, it uses a UV-curable adhesive. After grinding, the tape is exposed to ultraviolet light so it can be peeled off. The video notes that in mass production, applying and removing this tape is generally fully automated.

The numbers explain why this step is needed. A wafer cut from an ingot is about 725 µm thick for an 8-inch wafer. The target thickness depends on the type of semiconductor. For power semiconductors it is around 100 µm, and memory chips and similar devices are ground even thinner. The thinner the wafer, the higher the risk of chip breakage, so handling requires great care.

Once the wafer is loaded, the machine runs automatically through three stages: rough grinding to get close to the target dimension, main grinding to reach the exact thickness, and cleaning. Processing time depends on conditions, but it usually takes only a few dozen minutes. The camera looks inside the machine at the grinding stone that does the work. The wafer comes out cleaned, and it is visibly thin and fragile.

The host observes that although the step is called "grinding," it is closer to polishing, and the resulting surface is quite smooth. They explain that this surface roughness matters later: it affects how well the chip bonds to the lead frame and how reliable the finished part will be.

Swapping Tapes Before Dicing

Before dicing, the wafer must be mounted on a different tape, called dicing tape. First, the backgrinding tape is exposed to UV light in a dedicated unit to weaken its adhesion. Then the same tape-application machine used earlier, now loaded with dicing tape, applies the dicing tape to the wafer. Finally, the backgrinding tape is peeled off. Because the UV exposure has weakened it, the tape comes off cleanly.

4:39

Dicing and the Problem of Chipping

Dicing cuts the wafer into individual chips, called dies. The dicing saw uses an ultra-thin blade made with synthetic diamond abrasive grains. The blade spins at high speed and is pressed onto the wafer, cutting in straight lines while water is sprayed on it.

The main concern during dicing is chipping, where small chips and cracks form along the edges of the silicon die. According to the video, chipping can degrade the finished semiconductor's electrical characteristics and reduce its mechanical strength, which seriously affects reliability. In mass production it lowers yield. Chipping is linked to the grain size of the blade, the cutting conditions, and the tape. Because silicon is brittle, even small factors can cause it. The video shows an actual example of a chipped edge. For this reason, setting parameters such as blade speed and feed rate takes a great deal of expertise. After dicing, the separated chips remain stuck to the dicing tape.

7:11

Die Attach: Mounting the Chip on the Lead Frame

The next step removes each chip from the dicing tape and mounts it on a lead frame or substrate. This is called die attach or die bonding. The host holds up a TO-247 lead frame, a shape many viewers will recognize from the legs of power semiconductors. The chip sits on a flat area called the die pad. Depending on the application, it is joined with sheet solder, paste solder, sintered materials, and so on. In most power semiconductors, the drain or collector connects to the die pad.

A bare diced wafer will not take solder. So after backgrinding, metal is deposited on the back side for solder connection. This layer is called back metal. Die bonding comes down to stacking three things in order, the lead frame, the solder, and the chip with its back metal, and then heating them.

For paste solder, the video first shows a manual stencil printer, which suits high-mix, low-volume prototyping. Solder paste is spread over a stencil with a spatula to print it onto the lead frame. The host points out that this is nearly the same as solder paste printing for printed circuit boards. For filming purposes, this step was done outside the clean room.

The chip is then placed with a die bonder, which picks up a diced chip with its tip and sets it on the substrate or lead frame. The demonstration uses a different package from the TO-247. The chip can also be placed by hand onto the pre-printed paste, but that requires very careful work.

9:27

Reflow and Why Voids Matter

The assembly then goes through a reflow oven, which melts the solder and bonds the chip to the die pad. The oven runs in a nitrogen atmosphere, which the host again compares to PCB assembly. The chips come out well attached, but flux residue remains. If it is left, it can cause defects in later steps, so it has to be cleaned off.

Before moving on to cleaning, the video takes a closer look at the reflow step. When paste solder containing flux is used, cavities with no solder can form in the chip joint. These are called voids. They form because the solvents in the flux evaporate when heated, producing gas that gets trapped in the molten solder. Voids shrink the heat path from the chip to the lead frame, which hurts heat dissipation and also weakens the joint, lowering reliability.

The obvious question is why not just skip the flux. The video's answer is that without flux, the oxide film on the metal surfaces cannot be removed, and the bond between lead frame and chip would be weak.

11:09

Formic Acid Reflow: Soldering Without Flux

The solution presented is formic acid reflow, a fluxless soldering method. The host says they cannot go into detail here and will explain it elsewhere. Inside the clean room, the video shows that formic acid is used to reduce oxide films and oxidized metal, which removes the need for flux. The solder also changes from paste to pellet-form solder. According to the video, this eliminates flux-derived voids entirely and reduces void formation overall.

This is especially important for power semiconductors, which need to move the chip's heat efficiently into the lead frame. Voids undermine that, so joints with as few voids as possible are required. In the demonstration, the parts are heated in a special device that keeps a formic acid atmosphere. The finished result shows solder that has wetted and spread well, as in ordinary reflow, but with no flux residue.

12:07

Flux Cleaning and Rinsing

Back in the flux-based flow, the flux has to be removed. There are many ways to do this. Here, a cleaning solution is used. The reflowed lead frames are set in a jig and immersed in chemical solution in a cleaning unit. How well the parts get cleaned depends on the type of chemical, soaking time, and temperature, and the process takes roughly a few minutes to a few hours.

Next comes rinsing, which removes the cleaning chemical. The parts are dunked into a first rinse tank, drained thoroughly, and moved to a second rinse tank. After the second rinse and drying, they are ready for the next step.

13:30

Wire Bonding: Ball Versus Wedge

Wire bonding makes the electrical connections between the chip and the external electrodes. Die attach connects only the back side of the chip, so wires are needed for the remaining gate terminal and the source or emitter terminal.

The video describes two methods. Ball bonding uses gold, silver, or copper wire a few tens of microns in diameter and is often used for logic ICs. Wedge bonding usually uses aluminum or copper wire, which is much thicker. The thickest is about 500 µm. Because large currents flow through power semiconductors, they use thick wire and therefore wedge bonding.

In the wedge bonder shown, the aluminum wire is joined by applying load and ultrasonic energy, connecting the chip electrodes to the lead frame. For filming, the machine ran at a much slower speed setting. The host notes that mass-production wedge bonders reportedly run faster. With the wires in place, the part finally starts to look like a power semiconductor. After bonding, it gets a visual inspection before moving on to molding. The tour also shows a ball bonder with gold wire, which runs much faster than the wedge bonder.

15:12

Plasma Cleaning Before Molding

Before molding, the parts are cleaned with plasma. This improves adhesion between the molding resin and the lead frame. Loading the parts takes great care, because a knock or drop can detach the bonding wires. Once loaded, the operator starts the cleaning with a button and waits for it to finish. The host remarks that the difference is hard to see with the naked eye.

16:13

Mold Encapsulation

In mold encapsulation, the chip is sealed in hardened epoxy resin or a similar material to shield it from the outside environment and give it durability and reliability. The familiar black body of the package is this mold material. It physically protects the chip and also provides electrical insulation.

There are two main resin sealing methods: transfer molding and compression molding. The equipment shown uses transfer molding. The wire-bonded lead frame is placed in the mold, the mold is closed, and heat-melted resin is injected. The raw material comes as pellets, which the host found interesting to see because mold compound is rarely seen in its raw state. The pellets are heated, melted, and forced under pressure into the cavity around the lead frame. The host notes that the video skipped many preparation steps, such as applying mold release agent to the mold, which are needed in practice. According to the video, real mass-produced power semiconductors are made by the same transfer molding method.

17:32

Laser Marking

The model number and manufacturer's mark on a power semiconductor's surface are laser-engraved. In mass production this is done at this stage. For prototypes, it is sometimes done after the parts are separated. For this tour, the parts were marked "Ichiken x Shiima Electronics."

Tie Bar Cutting and Singulation

The last shaping step is tie bar cutting, which trims away the excess parts of the lead frame. The frame is placed in a die on a hydraulic press and cut. What comes out is the power semiconductor shape everyone recognizes.

18:09

Inspection and Yield

For mass-produced products, the video explains, inspection follows. It mainly consists of visual and electrical testing. Visual inspection looks for cosmetic defects. For automotive customers, or others with strict appearance requirements, every part is inspected. For consumer products, spot checks are sometimes enough. Electrical testing checks that the device works properly, and every unit is tested, even for consumer products.

On yield, the video reports that the back-end process for power semiconductors achieves quite good yields, in the high 90s and close to 100%.

Closing

The host sums up the tour as a chance to see the back-end process openly, which is not possible at a mass-production factory. They point viewers to Shiima Electronics and the earlier front-end video. The host also mentions that requests to film factory tours and work-experience videos for recruiting have been growing, and says the channel welcomes such projects.