Thermal steps do not make headlines, but they anchor the process flows of power devices, sensors and specialty CMOS: gate and field oxides, dopant drives, densification anneals, LPCVD polysilicon and nitride. When a fab expands or a line is rebuilt, the furnace decision returns — and with it the vertical-versus-horizontal question.

Both architectures do the same physics. The differences are in uniformity mechanics, contamination behaviour, floor-space economics and automation. Here is the framework we use.

Where each architecture starts ahead

Vertical furnaces dominate modern installations for reasons that compound at volume:

  • Load uniformity. Wafers sit horizontally in a vertically-stacked boat; gravity acts symmetrically, and gas flow can be engineered around a rotationally symmetric load. Temperature and film uniformity across large-diameter wafers generally benefit.
  • Contamination and ambient control. A vertical tube seals against a smaller opening; loadlock and inert-ambient options are more natural, which matters as device sensitivity rises.
  • Automation. Vertical platforms integrate cleanly with cassette or FOUP handling and factory automation — the default for 200 mm and above.
  • Footprint. Floor space per wafer processed is usually better.

Horizontal furnaces keep genuine advantages in specific contexts:

  • Capital cost and simplicity. For smaller wafers, R&D lines and less automated environments, a horizontal stack of tubes is economical and maintainable with modest infrastructure.
  • Flexibility for small batches and research. Loading is direct; changing boats, tubes and processes is straightforward for a multi-user lab.
  • Legacy continuity. A line already running qualified horizontal processes carries requalification cost with any architecture change — sometimes decisive on its own.

The decision inputs that matter

Architecture follows the application. Pin these down first:

  1. Wafer size and roadmap. At 150 mm the choice is open; at 200 mm and above, automation and uniformity push strongly vertical. If a 150→200 mm migration is plausible within the tool’s life, buy for the destination.
  2. Recipe mix. List the actual steps — oxidation (dry/wet), drives, anneals, LPCVD films — with temperatures and durations. A furnace bank is sized against a recipe portfolio, not a single process.
  3. Device sensitivity. Metallic-contamination and moisture requirements decide the ambient-control features you must have, which are architecture-linked.
  4. Throughput model. Batch size × cycle time against your real lot arrival pattern. Oversized batches processed half-full destroy the cost advantage that justified batch processing.
  5. Facility reality. Ceiling height, floor loading, gas and abatement provisions, and the automation your fab actually runs.

Evaluation data that settles it

Whatever the architecture, ask every candidate for the same evidence:

  • Temperature uniformity across a full load, at your process temperatures — not an empty-tube profile;
  • Film thickness uniformity (for oxidation/LPCVD) wafer-to-wafer through the load, including the boat ends;
  • Ramp rates achievable within your thermal budget, and recovery behaviour after boat push;
  • Particle data at your film and temperature, over a stated interval;
  • Metallic contamination performance appropriate to your device class;
  • Maintenance profile: tube and boat replacement cycles, and what each costs in downtime.

Cost per wafer pass, honestly computed

The batch furnace’s reason for existing is economics, so compute it properly: amortised capital + energy + gases + consumables (tubes, boats) + maintenance labour, divided by realistic annual wafer passes — using your actual expected batch fill, not the brochure’s. This number, next to the uniformity data, usually makes the architecture decision for you.

One caution

Furnace decisions have long shadows: the tool will likely outlive the product that justified it. Weight recipe flexibility and upgrade paths accordingly — the cheapest furnace for this year’s process is not always the cheapest furnace for the decade.

Evaluating batch thermal capacity for a power, sensor or specialty line? Talk to us about your recipe portfolio.