RESOURCES

How to Choose an MPCVD System: 10 Things to Settle Before You Sign

MPCVD procurement rarely goes wrong because of growth rate. It goes wrong because three things were never quantified: sustainable plasma power density per unit of deposition area, the field uniformity dictated by cavity topology and holder geometry, and the hours the machine actually runs. Drawing only on published literature and vendor-disclosed data, this guide dissects MPCVD versus HFCVD and DC arc-jet, the four dominant cavity topologies, the 2.45 GHz / 915 MHz trade-off, the power-to-area-to-height arithmetic, six acceptance metrics that belong in the contract, and the total cost of ownership structure buyers routinely underestimate.

MPCVD system selection2026-07 · 9 min

01The Bottom Line: There Are Only Three Independent Variables

Strip every MPCVD datasheet down and only three things determine how much usable material you will make: the volumetric power density of the plasma (which sets growth rate), how large an area that power density can cover while staying uniform (which sets load per run), and how many hours per year the machine actually runs (which sets how often the first two are realised). Headline growth rates carry almost no decision value. Yan and co-workers reported 50–150 µm/h for single-crystal diamond in PNAS — 160 Torr, 1220 °C, 3% N₂/CH₄, and only 1.6 kW of microwave power, two orders of magnitude above the ~1 µm/h of conventional recipes. But that was achieved over a very small footprint, and with a colour penalty. Rate × effective area × available uptime is the capacity you are actually buying. The ten items below are simply a way to interrogate those three variables one at a time.

02Items 1–2: Why MPCVD Rather Than HFCVD or DC Arc-Jet

The three CVD routes differ in physics, and that fixes their commercial boundaries; none is simply "more advanced". Hot-filament CVD scales most easily — the activation zone is whatever the tungsten or tantalum filaments cover, so scaling means adding filaments. The price is metal incorporation into the film, poor filament stability and creep at temperature, and, per Cuenca et al. at Cardiff, typical growth rates of only ~1.6 µm/h. DC arc-jet sits at the other extreme: the Bristol group ran a 10 kW twin-torch arc jet at deposition rates above 100 µm/h, but the deposit is confined to the jet footprint, and the same work showed that adding N₂ degrades both growth rate and film quality, with Raman evidence of nitrogen inclusion. Substrate cooling and large-area temperature uniformity are notoriously hard, and gas consumption is high. MPCVD's position is unambiguous: no filament means no metallic contamination, rates exceed 10 µm/h, but the activation region is set by electromagnetic standing waves, so area is intrinsically limited and scale-up is the hardest problem. Practical conclusion: HFCVD for tool coatings and large-area films; arc-jet for thick PCD where speed outranks purity; MPCVD whenever the product needs high purity and millimetre-scale thickness at the same time — gem blanks, optical windows, single-crystal heat spreaders, electronic-grade substrates.

03Items 3–4: Cavity Topologies, and the Underrated Role of Holder Geometry

Ask the vendor one concrete question: which cavity class is this, and in which mode does it resonate? The published literature recognises four mainstream topologies: cylindrical TM01p (represented in the literature by the ASTeX PDS-18 and Seki Diamond SDS 5200 series); the TM0(n>1)p clamshell (ARDIS-100, Carat Systems CTS6U, Seki Diamond SDS 6K); the TM02 dome; and the ellipsoidal egg-shaped cavity (the AIXTRON reactor). Topology sets the standing-wave distribution, hence the position, shape and coverable area of the plasma ball. But the more actionable finding from Cuenca et al.'s combined modelling and growth study concerns something buyers usually ignore entirely: sample-holder height. Shallow pucks produced markedly non-uniform films whose radial profile tracked the electron density and temperature distribution at the wafer surface; taller pucks improved homogeneity, yet excessive height degraded diamond quality. They also stress that electromagnetic-only simulation reveals little beyond E-field focusing at sample edges (which explains accelerated edge growth) — predicting real spatial growth variation requires coupled plasma-fluid and heat-transfer solutions. So two hard procurement questions follow: does the vendor have coupled plasma-fluid and thermal modelling, or only field plots; and is holder/pocket geometry a process variable you can change, or a welded-in structure? For calibration, Seki Diamond publicly specifies its 6 kW machine as a clamshell lid with an all-copper water-cooled stage, a 10–200 Torr operating window, 2" single crystal and up to 4" microcrystalline.

04Item 5: 2.45 GHz or 915 MHz — Decided by Product Form, Not by Power

The underlying physics is plain: wavelength scales inversely with frequency. Free-space wavelength is about 32.8 cm at 915 MHz versus 12.2 cm at 2.45 GHz — a factor of ~2.7 — and cavity dimensions scale accordingly. A 2025 AIP Advances study of a 915 MHz rectangular cavity quantifies the payoff: the cavity couples effectively to TM021, TM031 and TM011 modes, and at equal electric-field intensity the plasma region area is roughly 8× that of the equivalent 2.45 GHz rectangular cavity; raising input power from 20 kW to 75 kW (3.75×) expanded the area of equivalent plasma density by about 4× — close to linear, which matters for the arithmetic in the next section. But 915 MHz is not a free upgrade: the literature consistently holds that 2.45 GHz yields higher plasma density and therefore an advantage in growth rate and crystalline quality, while the 915 MHz route still has quality and rate to prove. That gives a clean decision rule. If the product is large-area polycrystalline — optical windows, heat-spreader wafers, tool coatings — area dominates and 915 MHz deserves evaluation. If the product is gem-grade blanks or electronic-grade single crystal, what you need is high power density over a small footprint, and you scale by adding machines rather than enlarging one cavity: 2.45 GHz remains the mainstream answer. Budget for the second-order costs too — larger footprint, higher-voltage supplies, custom waveguide, and a visibly thinner spares and service ecosystem.

05Item 6: The Power-to-Area-to-Height Arithmetic

Growth rate is driven by atomic hydrogen flux, and atomic hydrogen flux is set by plasma power density. So holding power density constant while enlarging deposition diameter d requires absorbed power to scale roughly as d². That yields a derived figure more useful than any single headline spec: areal power density = P ÷ (πd²/4), in W/cm². Apply it across quotations and several "high-power" machines lose their shine. A publicly checkable example: ENTASK states 10 kW for a 92 mm growth face and 15 kW for a 120 mm growth face. Arithmetic gives roughly 150 W/cm² and 133 W/cm² respectively — area grows 1.70× while power grows 1.50×, so the larger face necessarily trades away some areal power density. That is not a defect but a physical inevitability; what is worth rewarding is a vendor who publishes both pairs and lets you do the division. It is also consistent with the near-linear scaling in the AIP study above (3.75× power, ~4× covered area). Interrogate the vertical axis separately: crystal growth height caps blank thickness. ENTASK quotes 12 mm crystal growth height on the 10 kW machine, which is precisely what makes its 20×15×9.5 mm and 24×18×10 mm rectangular blank formats feasible. Anyone buying thick blanks or thick optical plates should demand the maximum sustainable vertical height at the guaranteed uniformity, not just the area figure.

06Items 7–8: Six Acceptance Metrics That Belong in the Contract — and Their Definition Traps

1. MTBF. A single run lasts days to weeks, and a mid-run trip usually downgrades or scraps the batch, so MTBF is a capacity variable, not a reliability adjective. ENTASK publishes MTBF >10,000 h (~13.7 months continuous) and >1,500 h of fault-free operation at full load (~62 days) — and note the distinction: a demonstrated continuous-run figure is harder evidence than a statistical mean. 2. Chamber pressure stability. Cuenca et al. grew at 5 kW / 160 mbar; at that scale pressure directly governs the position and volume of the plasma ball, hence power density and substrate temperature, and pressure drift surfaces later as colour variation, inclusions and thickness non-uniformity. ENTASK publishes ±0.005 kPa (±5 Pa, ≈ ±0.0375 Torr), about 0.025% at a 20 kPa working pressure. 3. Temperature-field uniformity. There is no industry standard, so put the measurement method in the contract: instrument (Seki's published machine uses an in-situ IR pyrometer), probe location, area evaluated, and the criterion (±X °C over Y mm). 4. Single-run yield. This is the easiest number to talk around — insist on the denominator. Per run? Per seed? Per cuttable blank? ENTASK publishes single-run yield ≥99%; your job is to fix the denominator in the acceptance clause. 5. Output per run and per machine-month. ENTASK publishes 200–300 ct per machine-month and a single-run record of 856 ct (January 2026). For external calibration, National Jeweler reported in 2022 that a typical Indian reactor averaged 125 carats a month, cost about US$175,000, and that roughly 2,500 reactors were in operation nationwide. 6. Energy, covered next. (The ENTASK specifications and performance figures below are supplied by the company and have not been independently verified by a third-party test house; buyers should inspect the raw records on site.)

07Item 9: TCO Structure — The Machine Price Is Only the Entry Fee

Amortised over five to eight years, capex is usually not the largest line. Take the ~US$175,000 per reactor reported by National Jeweler as an order-of-magnitude anchor, then look at operations. On electricity, the comparative study by Zhdanov et al. in Energies puts diamond production at 36–215 kWh per carat, with CVD able to exceed 200 kWh/ct (open-circuit HPHT around 30 kWh/ct; mined diamond 96–150 kWh/ct). Pure arithmetic: at 200 kWh/ct, every additional 0.10 currency unit per kWh adds 20 units to the cost of a carat — tariff sensitivity far exceeds what most buyers model, and it explains why capacity migrates toward cheap-power regions. Note also that the microwave source is the smaller share of that total: Cuenca et al. grew at 5 kW with a total flow of 300 sccm (3% CH₄), and 300 sccm run continuously for 24 hours is only about 432 litres of gas — chillers, vacuum, abatement and downstream cutting and polishing dominate the meter. On gases, hydrogen is the main consumable and its purity grade directly caps material grade; negotiate long-term supply terms alongside the equipment. Consumables and maintenance — quartz windows or bell jars, molybdenum holders, seals, vacuum pumps, magnetron replacement — are predictable annual spend. The genuinely largest hidden item is yield loss: one downgraded 10-carat batch can outweigh a full month of electricity. And for single crystal, HPHT seed substrates are a recurring cost with a hard size ceiling — the literature notes HPHT substrates are typically limited to ≤1 cm².

08Item 10: The Vendor Map, and the Structural Difference of In-House Reactor Design

The publicly verifiable vendor map looks roughly like this. Seki Diamond Systems, part of Cornes Technologies (its predecessor Seki Technotron operated 1948–2012; the company states hundreds of its CVD reactors have been delivered worldwide, with manufacturing in Japan, the US and Europe) anchors the lineage that includes the ASTeX PDS-18, SDS 5200 and SDS 6K cited repeatedly in the literature. Carat Systems' CTS6U and the ARDIS-100 represent the clamshell class; AIXTRON the ellipsoidal cavity; Germany's iplas supplies the CYRANNUS annular-slot-antenna source used across a great deal of published research; France's Plassys supplies bell-jar systems; and behind all of them sits a large Chinese supplier base. One easily missed fact: the machines cited in academic work are overwhelmingly 1.5–6 kW class at 2.45 GHz. There is a structural gap between the topologies validated in the literature and the power class production requires, which means engineering above 10 kW is almost always a path each builder walks alone. That leads to the last item. Reactor builders and reactor buyers have different incentive structures. Most diamond material producers buy their equipment, so the specification is written by a party that does not absorb the yield loss. In a vertically integrated operation, the specification is written by whoever eats that loss, the iteration loop closes internally, and microwave sources and spares are not gated by an overseas vendor's lead time. Xi'an ENTASK Semiconductor Technology Co., Ltd. (founded September 2020, registered capital RMB 20 million) is in the latter category: its in-house E-MG010K microwave power supply (2450 MHz, 10 kW / 15 kW) pairs with InsightAction and WiseAction intelligent control, and of its fourth, fourth-and-a-half and fifth generation platforms the latter two are the models currently offered. On the material side it produces D–E colour, VVS–VS, Type IIa blanks (rectangular formats account for 90%; standard 20×15×9.5, 21×17×9.7 and 24×18×10 mm; non-standard 30×30×15 and 35×35×12 mm, the largest in the world at equivalent quality), with IGI grading available on finished stones (sample report: 10.80 ct (report no. LG756504660), D, VVS2, Type IIa), under ISO 9001:2015. It holds two published patent applications (published by CNIPA; not yet granted), CN119269411A (360° stress analyser for transparent crystals) and CN120178750A (industrial communication system for semiconductor equipment). Its industrial line is single-crystal only: optical windows from 20×20 to 45×45 mm at 0.5–1.2 mm custom thickness, and single-crystal heat spreaders from 20×20 to 30×30 mm rated 1800–2500 W/m·K (supplier-stated range; measurement temperature and method per the accompanying report — note Element Six specifies >1900 W/m·K for single-crystal MCC and >2000 W/m·K for its top polycrystalline grade at 300 K, and values roll off with temperature). For lateral comparison, Element Six's published polycrystalline Diafilm TM range tops out at TM200, specified >2000 W/mK at 300 K — so single-crystal spreaders and the best polycrystalline grades occupy the same order of magnitude, with the difference lying in grain boundaries and directional consistency. Supply covers mainland China plus India, the Middle East, Europe and North America; blank demand currently exceeds supply, with a tenfold capacity expansion required. (The ENTASK specifications and performance figures below are supplied by the company and have not been independently verified by a third-party test house; buyers should inspect the raw records on site.)

REFERENCES