At a glance
A close look at the five bodies of work behind deep reactive ion etching, from IBM's ion-beam experiments to the Bosch cycle, cryogenic etching and the black silicon method.
A process engineer once asked me, half joking, why the Bosch process carries a German name when the physics underneath it was worked out in California. It's a fair question, and the honest answer takes in four countries and roughly thirty years.
Deep reactive ion etching (DRIE) is the plasma process that cuts near-vertical holes and trenches into silicon, often tens to hundreds of micrometres deep. It's behind the accelerometer that fires an airbag, the microphone in a phone that hears you across a room, and the copper-filled vias that connect stacked memory dies through the wafer. The technique rests on one piece of physics and three separate engineering solutions built on top of it, and each of those solutions has a small team behind it.
What follows is five bodies of work rather than five lone inventors, because three of them were partnerships. That isn't a hedge. It's how this corner of process science got done: a physicist and a surface chemist sharing a lab, a process engineer and a colleague who knew surface engineering, a group of instrument builders who wrote a procedure instead of a patent.
The mechanism: why ions and fluorine together cut straight down
Silicon etches readily in fluorine, but it etches in every direction at once, undercutting the mask and rounding the profile. Ions alone will sputter silicon, slowly and without chemical help. John W. Coburn and Harold F. Winters, working at IBM's research laboratory in San Jose, showed in 1979 that the two together do something neither does on its own: with ions striking the surface while the fluorine chemistry proceeds, the etch rate climbs well past the sum of the separate rates.
The experiment that made the point was deliberately plain. They etched silicon with xenon difluoride outside a plasma altogether, then aimed a beam of ions at it. Where the ions landed, etching accelerated. Where they didn't, very little happened. Directionality turned out not to be a property of the gas but of where the energy was delivered.
Winters stayed on IBM's research staff until he retired in 1993, and the pair's later review of etching surface science became the reference point for anyone trying to model the process rather than tune it by trial. The American Vacuum Society still gives a student award in both their names, which is a more durable kind of citation than most.
Cold silicon: Tachi at Hitachi, 1988
Shinichi Tachi's team at Hitachi's Central Research Laboratory in Kokubunji attacked the same problem from the thermal end. Their 1988 paper in Applied Physics Letters described etching silicon in sulphur hexafluoride while holding the wafer between minus 130 and minus 100 degrees Celsius. Sideways etching all but stopped. Selectivity to an organic resist reached 30:1, with rates of 500 nanometres a minute in reactive ion etching and 1,000 in a microwave plasma source.
The reason matters, because it separates the two DRIE families in use now. At those temperatures a silicon oxyfluoride layer forms on the sidewalls and blocks the fluorine radicals, and most of that layer leaves when the wafer warms back up. No fluorocarbon polymer is deposited, so the chamber doesn't need the clean-up cycle a Bosch tool does. What you pay for is plumbing and patience: liquid nitrogen cooling, careful thermal contact, and a window narrow enough that a few degrees of drift will show up in the profile. The 1988 paper is worth reading directly, and it drew a formal comment and reply in the same journal before the year was out.
Not everyone accepted the explanation. The comment disputed the mechanism; Tachi's reply argued from the manufacturing side, that room-temperature SF6 etching forced a choice between a vertical wall and a mask that survives, and that the cold process removed the choice. The exchange is a useful reminder of how quickly an argument about surface chemistry becomes an argument about yield.
The alternating cycle: Lärmer and Urban at Bosch
Franz Lärmer joined Bosch's corporate research centre in 1990, a physicist with a doctorate from the Technical University of Munich and, by his own description, a strong affinity for chemistry. He has said his manager handed him a task the literature treated as impossible. With Andrea Urban, then publishing under her maiden name Schilp and trained in surface engineering at Aalen, he stopped trying to win the fight between chemistry and directionality and split the process in time instead.
The cycle runs in most DRIE tools today, and it takes four lines to describe.
- Etch: a sulphur hexafluoride plasma releases fluorine that attacks silicon downwards, and sideways.
- Passivate: a fluorocarbon plasma, usually from C4F8, lays a polymer film over everything, walls included.
- Break through: the substrate bias pulls ions straight down, and they clear the polymer from the trench floor before they clear it from the walls.
- Repeat: hundreds of times, each cycle removing a shallow slice of silicon and rebuilding the wall protection behind it.
The patent was filed through the PCT in November 1993 and granted in the United States in March 1996, with a German priority date of December 1992. Performance moved quickly: etch rates above three micrometres a minute with mask selectivity beyond 100:1 in the patent's early description, and close to ten micrometres a minute with selectivity above 300:1 by the team's 1999 conference paper.
Commercialisation ran through a toolmaker rather than a fab. Surface Technology Systems, then based in the United Kingdom, worked with Bosch to bring the process to market and became its first licensee, which is the main reason the company's name turned into shorthand: fabs bought the recipe inside a machine instead of inventing it. Bosch shipped the process in an airbag in 1997, then in new gyroscope generations in 1998 and 2000. By 2007 the company was reporting 50 million acceleration and angular-rate sensors a year built on the technology. The Eduard Rhein Foundation, which gave Urban and Lärmer its technology prize in 2019, counts more than ten billion MEMS devices out of Bosch since 1995 and notes a new microsystems fab under construction in Dresden.
The awards arrived in a cluster: European Inventor of the Year in the industry category for 2007, the IEEE's Jun-ichi Nishizawa Medal for Lärmer in 2014, and the Eduard Rhein prize in 2019 shared by both. Lärmer holds more than 150 patents and is a Bosch Research Fellow. Urban's name is on the original patent, and it's worth saying plainly, because conference programmes and press coverage have a habit of compressing a two-person invention into a company adjective.
Photo by Marwen Larafa on Unsplash
The method: Jansen and the black silicon procedure at Twente
Henri Jansen's contribution is a procedure rather than a process, and it solved a problem every lab with a fluorine etcher recognised. With Meint de Boer, R. Legtenberg and Miko Elwenspoek at the MESA Research Institute in Twente, he published the black silicon method in 1995: a systematic way of finding the parameter settings that produce a vertical wall on a given tool, instead of guessing at power, pressure and gas ratio. They demonstrated trenches 200 micrometres deep at aspect ratios up to 10 in SF6/O2/CHF3 chemistry, and validated the procedure on three different reactors, two parallel-plate and one hexode.
Fifteen years later the same series of papers took on a problem the Bosch cycle had created for itself. C4F8 is a hydrofluorocarbon, and restrictions on HFCs were tightening. Jansen's group replaced the passivation pulses with oxygen, held the chuck between minus 120 and minus 80 degrees Celsius, and reported silicon etch rates up to 25 micrometres a minute while keeping profiles vertical. Same cold-wall chemistry Tachi had found in 1988, considerably better controlled.
The map: Wu, Kumar and Pamarthy at Applied Materials
By 2010 the field had enough history to need a map. Banqiu Wu, Ajay Kumar and Sharma Pamarthy, working at Applied Materials, published a review in the Journal of Applied Physics that traced the line from wet etching in solution through low-density reactive ion etching, cryogenic single-step etching in an inductively coupled plasma, and the time-multiplexed cycle, to single-step high-density etching near room temperature. They also said the quiet part out loud: the multiplexed process was the industrial workhorse, and its scalloped sidewalls were intrinsic to switching gases rather than a defect anyone would ever fully design away.
Their review named the metrics fabs still argue about — aspect ratio, sidewall profile, aspect-ratio-dependent etching, loading — and pointed at the application that would pay for the next round of development: through-silicon vias for stacking chips in three dimensions. That framing held. TSV work is why high-aspect-ratio silicon etch stopped being a MEMS specialty and became a packaging requirement.
Four countries, four ways of getting the same idea into factories
Set the four origins side by side and the institutional logic is visible. IBM's contribution was mechanistic and came out of a corporate basic-research laboratory, which is why roughly fifteen years passed before it reached production floors. Hitachi's came from a central laboratory with a manufacturing brief and reads like a process paper from the first page. Bosch's came from an industrial group that patented in 1992 and licensed to a toolmaker, which compressed the gap between invention and product to about five years. Twente's came out of an instrument-building group and produced a method — less commercially valuable, considerably more widely used.
None of these routes is the correct one, and the lesson for anyone assembling a process team now is that the route determines what you end up holding: a patent and a licence, a method paper, or a mechanism nobody outside the lab understands for a decade. All three finished up in the same tools. The 1979 Coburn and Winters paper remains the cleanest short statement of ion-enhanced etching, and it rewards a direct read rather than a summary.
What the next decade is asking of this toolkit
DRIE now carries loads its inventors didn't design for. Consumer microphones and inertial sensors run in volumes that make automotive volumes look modest. Through-silicon vias in stacked memory and interposers demand uniformity across full 300-millimetre wafers, not just a good profile in the centre. At the depths those parts need, the limiting factor is often transport — how many radicals and ions actually reach the bottom of a narrow hole — rather than the surface chemistry Coburn and Winters measured. The 2025 reviews of deep silicon etching spend most of their length on uniformity and aspect-ratio effects for exactly that reason.
Research attention has moved toward atomic layer etching, which uses the same ion-enhanced surface chemistry but self-limits each cycle to a fraction of a layer; low-temperature versions have already been demonstrated for silicon dioxide. The alternating-cycle idea Urban and Lärmer patented has, in a sense, been generalised into a control philosophy. And the descendants of the black silicon method are still the fastest way to bring a new etcher into specification.
Photo by Vitaly Gariev on Unsplash
One hour worth booking before the mask is written
If your team is specifying a deep silicon etch this quarter, there's a change that costs nothing and prevents most of the pain: get the process engineer into the design review before the mask is written. Aspect-ratio-dependent etching means the width you draw sets the depth you actually get, and no lithography step downstream repairs that decision. Jansen's 1995 paper was, underneath the parameter tables, an argument that profile is something you choose rather than something you accept.
The other habit worth copying from everyone on this list is duller. Write the procedure down. Tachi's reply to his critic, Jansen's method papers, the step-by-step language of the Bosch patent — all of them travelled further than the tools they were written for. A recipe that lives in one engineer's head retires when that engineer does.
- deep-reactive-ion-etching
- deep-reactive-ion-etching
- bosch-process
- bosch-process
- cryogenic-etching

