Guide

Compound semiconductors in India: two approved fabs, and the material argument behind them

Published 25 August 2026. Facts verified to 25 August 2026 unless dated otherwise.

The government's own count of its portfolio contains one silicon fab, one silicon carbide fab and one gallium nitride display fab, which is the cleanest way into this subject.

Start with the government's own categorisation

Compound semiconductor coverage in India tends to arrive as a list of materials with adjectives attached, which leaves a reader no way to tell an approved plant from a press release. There is a better starting point. When the Union Cabinet approved Semicon 2.0 on 15 July 2026 with a budget outlay of Rs 1,27,500 crore, the government also restated the manufacturing portfolio it had built to that date: twelve approved units carrying more than Rs 1.64 lakh crore, made up of one silicon fab, one silicon carbide fab, one integrated gallium nitride micro light emitting diode display fab and nine packaging units DURABLE. That single sentence sorts the field. Two of the twelve are compound fabs, one is silicon, and the rest are back-end plants.

This page takes that categorisation as its spine and fills in each entry with dated, sourced detail. It also explains what a compound semiconductor is in industrial rather than promotional terms, because the material difference is what drives every cost, capacity and timeline figure that follows. The engineering of wide-bandgap devices is treated separately on the fab desk; this file is about projects, capital and the honest position of India inside a global business that has just been through a painful cycle.

What a compound semiconductor actually is, and why it costs what it costs

Silicon is a single element. A compound semiconductor is made from two or more, and the useful families here are silicon carbide, which pairs silicon with carbon, and gallium nitride, which pairs gallium with nitrogen. Both have a wider bandgap than silicon, which in practical terms means they tolerate much higher electric fields. A device that would need a thick, resistive drift region in silicon to block 650 or 1,200 volts can do the same job in a far thinner layer of silicon carbide, so it conducts with lower losses and switches faster. Gallium nitride does something similar at lower voltages and higher frequencies, and it also emits light efficiently, which is why the same material underpins light emitting diodes and micro light emitting diode displays.

The costs follow from the materials rather than from lithography. Silicon carbide boules are grown slowly at very high temperatures and are extremely hard to slice and polish, so substrates have historically been the industry's bottleneck and its largest single cost. Compound device performance depends on epitaxy, the growth of precisely doped crystalline layers on that substrate, which is a specialist step performed in dedicated reactors. Wafer diameters are smaller than in silicon logic, with 150mm the volume standard for silicon carbide and 200mm still in transition, and gallium nitride on silicon carbide or silicon commonly running at 150mm or below. Smaller wafers, slower growth and harder materials together mean the cost per finished die stays high, and no lithographic advance fixes any of it.

That is why compound fabs are not miniature versions of silicon fabs. They are cheaper in absolute capital because the wafers are smaller and the tool set is different, and they are structurally more exposed to substrate supply, epitaxy know-how and yield learning than to node race dynamics.

SiCSem at Bhubaneswar: India's first silicon carbide fab

The silicon carbide entry in the portfolio is SiCSem, approved under the India Semiconductor Mission in August 2025 within the batch of four units cleared by the Union Cabinet on 12 August 2025 for Odisha, Punjab and Andhra Pradesh at an outlay of about Rs 4,600 crore. The government figures for the project are about Rs 2,066 crore of approved investment, 5,000 silicon carbide wafers a month and a packaging line of 8 million units a month DURABLE. The project is a partnership with Clas-SiC Wafer Fab of the United Kingdom, which supplies the process technology.

Trade reporting adds the physical record. Compound Semiconductor and Evertiq both reported on 3 November 2025 that SiCSem, described as Chennai-based, had broken ground at Info Valley in Bhubaneswar on an end-to-end silicon carbide plant at about 220 million dollars, with 60,000 silicon carbide wafers a year and 96 million packaged units a year, targeting operations in 2027 or 2028 and claiming 5,000 direct and indirect jobs REPORTED. Those press capacities reconcile exactly with the government's approval: 5,000 wafers a month is 60,000 a year, and 8 million units a month is 96 million a year. When two independent descriptions of a project agree at that level of detail, the underlying numbers are usually real, which is a check this record runs wherever it can.

What the reconciliation does not settle is delivery. A groundbreaking is a beginning, the 2027 to 2028 operational window is a target rather than an outcome TARGET, and the employment figure mixes direct and indirect jobs, which are different things counted by different methods. The scoreable events ahead are the fiscal support agreement, evidence of epitaxy and process tool installation, and a first qualified device.

Crystal Matrix at Dholera: gallium nitride, and an open question

The gallium nitride entry sits inside the record this desk tracks most closely. On 5 May 2026 the Union Cabinet approved two further semiconductor units in Gujarat with a combined investment of more than Rs 3,900 crore and expected employment of 2,230 skilled professionals TARGET. One of the pair is Crystal Matrix Limited at Dholera, cleared for an integrated compound semiconductor fabrication and assembly, test, marking and packaging facility making mini and micro light emitting diode display modules, with gallium nitride foundry services including epitaxy on six inch wafers, at an annual capacity of 72,000 square metres of display panels and 24,000 sets of red, green and blue epitaxy wafers DURABLE. The government did not publish the split of the Rs 3,936 crore between the two projects, so this record does not invent one. The India Semiconductor Mission's own site, last updated 19 August 2026, lists the project as an integrated compound semiconductor fab and assembly plant at Dholera approved in May 2026.

The project also carries the most substantial published scepticism attached to any unit in the portfolio, and a record that prints inconvenient sourcing has to carry it. Business Today reported on 8 May 2026 that Crystal Matrix is Hyderabad-based, that its FY25 revenue was Rs 1.07 crore, that its disclosed fixed assets amounted to basic office equipment worth Rs 1.32 lakh with no semiconductor manufacturing machinery listed, and that Medha Servo Drives, a railway electronics firm without prior semiconductor experience, holds 26 percent of the company REPORTED. The same report noted the leadership the company has assembled, including a chairman who previously headed Bharat Electronics, an adviser with more than 35 years in semiconductor fabrication operations and a former Samsung executive advising the project, and quoted an industry analyst making the point that credibility and commercial readiness are not the same thing.

This desk takes no position on whether the project will deliver, because nothing in the public domain settles that. What can be said precisely is that the approval is real and dated, that the capacity and technology description is the government's own, and that the company's published financial base is small relative to the plant described. The same report placed the outcome on technology partners and yield learning, noted that India lacks domestic supply chains for the materials involved, and put the infrastructure requirement, including metalorganic chemical vapour deposition reactors and cleanrooms, in the hundreds to thousands of crore. That is a fair description of the technology: mass transfer, the step in which very large numbers of microscopic emitters are placed onto a backplane, is what has kept micro light emitting diode displays expensive wherever they have been attempted. The events that would move this from an open question to a record are ordinary and observable: a fiscal support agreement, a construction contract, metalorganic chemical vapour deposition reactor orders, and a named display customer. Our cluster map tracks what is physically present at Dholera against what is still on paper.

The compound work already inside silicon companies

Compound devices in India are not confined to the two approved fabs. Continental Device India at Mohali announced in September 2023 that it had become India's first manufacturer of silicon carbide components, alongside a new surface mount packaging line REPORTED, and its expansion approved in August 2025 at about Rs 117 crore explicitly covers high power discretes in both silicon and silicon carbide at roughly 158.38 million units a year DURABLE. That is device assembly and packaging using compound material rather than compound wafer fabrication, and the distinction matters: buying silicon carbide dies and packaging them is a different industrial capability from growing and processing silicon carbide wafers.

The same distinction runs through the rest of the portfolio. Several approved packaging plants will handle power and discrete devices whose dies may be compound, without any of those plants being compound fabs. Keeping the two apart is the difference between a supply chain that imports the hard part and one that makes it. The analog and power lane, including the demand logic that drives both, is covered in the analog and power file.

The global context India is entering, including the bad news

Compound semiconductors are a real industry with real cycles, and the recent cycle has been harsh. Wolfspeed, one of the largest specialists in silicon carbide, filed for Chapter 11 protection on 30 June 2025 and completed its restructuring on 29 September 2025, reducing total debt by about 70 percent, extending maturities to 2030 and lowering annual cash interest expense by roughly 60 percent according to the company's own announcement. The filing came after a stretch in which silicon carbide capacity across the industry ran ahead of electric vehicle demand growth, and substrate prices fell accordingly. On the materials side, DigiTimes reported on 14 July 2026 that six inch silicon carbide substrate prices, after roughly two years of oversupply and decline, had bottomed and begun to recover as capacity stayed constrained REPORTED. Device supply is concentrated among a small group of established manufacturers including STMicroelectronics, Infineon, onsemi, ROHM and Wolfspeed, all of which have been through the same price compression.

Three consequences follow for India. Entering a market after a price collapse is cheaper on equipment and harder on margins, since incumbents have already written down capacity and can defend share on price. A 5,000 wafer a month plant is a modest entrant by global standards, so its economics will depend on serving domestic and regional demand at qualified quality rather than on competing for global volume. And substrate supply, which was the industry's bottleneck through the shortage years, is now the industry's overhang, which is favourable for a buyer and unfavourable for anyone planning to grow crystals for export.

What is not public, and what to watch

Several things a reader might expect to find are simply not in the public domain as of 25 August 2026. Neither compound project has published a fiscal support agreement date. Neither has published a named customer. The split of the Rs 3,936 crore between Crystal Matrix and Suchi Semicon has not been released. No public document describes where either plant will source substrates, which for a silicon carbide fab is the most load-bearing supply question there is. And no compound project in India has published a qualification roadmap for automotive-grade devices, which is the standard that most silicon carbide value is sold against.

• Fiscal support agreements and disbursement. The gap between Cabinet approval and a signed agreement ran about twelve months for the Dholera silicon fab. Anything materially longer for these projects is a signal.
• Tool orders, especially epitaxy. Metalorganic chemical vapour deposition and silicon carbide epitaxy reactors are long-lead, specialised and hard to hide once ordered. They are the most reliable early evidence that a compound project is real.
• Substrate sourcing. Any announced supply agreement for silicon carbide or gallium nitride substrates would tell this record more about the projects' seriousness than another capacity headline.
• Semicon 2.0 materials support. The second phase covers materials as one of its stated areas. Whether any of it reaches compound substrates or epitaxy is a policy question with an observable answer.
• The count itself. Twelve is the current number, sourced to the Cabinet release of 5 May 2026 and unchanged on the mission's own site as of 19 August 2026. Our count answer carries the full portfolio, one line each, with the date stamp that makes it readable a year from now.

The compound lane is the smallest part of India's semiconductor programme by capital and the most technically distinct part by content. It deserves neither the dismissal it usually gets from logic-focused coverage nor the enthusiasm it gets from material-science boosterism. Two fabs are approved, and one of them has broken ground. The other holds an approval, a published capacity description and an unresolved question about the company behind it. That is the whole record on this lane, and it will be scored as it develops.