The fine and specialty chemicals business: how it really works: Inside the multipurpose batch plant

3.1 Small volumes and short product lives make the multipurpose plant the norm

All of the top 300 chemicals by volume are made in continuous plants. Further down it flips: 90% of those ranked 300 to 3,000 are made batchwise, and 97% beyond rank 3,000. Demand for a single custom fine chemical rarely exceeds 100 tonnes a year; a third of the top 500 drugs need 10-100 tonnes of API and only 7% more than 1,000 tonnes. Portfolios also turn over fast, so a dedicated plant may be obsolete before it pays back. Hence the multipurpose plant: one set of equipment running many products in turn, each in a 'campaign' of days to weeks, followed by cleaning and re-piping.

The working unit is the bay, or train: typically two or three jacketed, stirred reactors of up to 10 m3, one solid-liquid separator and one dryer. A jacket is an outer shell carrying steam, water or brine to heat or cool the contents. The separator is a centrifuge or a nutsche, a large filter vessel in which crystals are washed and pressed. Plants are stacked so gravity moves material down: charging at the top, then reaction, crystallisation, filtration, and drying and packing at ground level.

Standard conditions are -20 °C to 200 °C and 10 millibar to 3 bar, in stainless or glass-lined steel. Specials such as cryogenic reactors, 100-bar vessels or Hastelloy and tantalum are best installed only on firm demand, because excess flexibility costs money.

Pollak's rule of thumb is 15-30 tonnes of product per m3 of reactor per year for a one-step process, so a bay of 4-6 m3 vessels makes about 100 tonnes a year of one step, a typical API volume. The industry's median reactor is 4-6 m3. The flowchart follows one bay of a Lonza design study for Visp, Switzerland.

• Pollak (2011) · pdf pp. 71-82

3.2 A cGMP multipurpose plant costs close to $1 million per cubic metre

Because output cannot be stated in tonnes, the industry measures capacity in cubic metres of reactor and investment per cubic metre. Pollak's example is a US cGMP plant: two trains with six 4 m3 reactors (24 m3), two filters and two dryers, costing $21 million for equipment and installation, excluding building and land. That is $21 million / 24 m3 = about $0.9 million per m3. Plants for nonregulated intermediates cost much less; pharma companies typically spend two to four times as much; Indian and Chinese plants considerably less.

Size matters less than in commodities. Pollak says cost per cubic metre falls with reactor size to the power 0.5, so total cost scales with the square root of size. Six 2 m3 reactors would cost $21 million x (2/4)^0.5 = about $15 million, or $1.25 million per m3, about 40% more per unit of capacity than the 4 m3 design ($0.9 million), but not enough to justify one giant vessel.

The money goes mostly on flexibility and cleanliness, not vessels: piping and installation take a quarter or more. Piping can be fixed, with manifolds and coupling stations, for a narrow product mix, or process-specific, with reactors, pumps and filters connected campaign by campaign, often by hoses, for frequent changes and easier cleaning.

For cGMP work the plant must be 'qualified': documented proof that the design suits its purpose, that it was installed as designed, that it works across its operating range and that it makes product reproducibly. Control software must be validated and air handling designed against contamination. The result is capital intensity: Pollak notes that more than $1 of investment is typically needed per $1 of annual sales.

• Pollak (2011) · pdf pp. 73-77, 83, 97-98, 124

3.3 Raw materials are about a third of cost; most of the rest is the cost of owning the plant

Pollak's indicative cost structure shows how little of a fine chemical company's cost varies with output. Raw materials, including solvents, are about 30% and utilities only 4-5%. The rest is largely fixed in the short run: plant labour 10-15%, capital cost (depreciation and interest) 15%, plant overhead such as quality control, maintenance and waste disposal 10%, R&D 8%, marketing and sales 5%, and administration 15%. In discussing prices he puts raw materials at about 35% and conversion at about 55% of the selling price, together the cost of goods sold.

The contrast with commodities is sharp. Wittcoff notes that labour is a tiny share of petrochemical cash costs (2.5% for high-density polyethylene, 3.5% for benzene), while the 'fine chemicals' sectors, such as drugs and pesticides, are less capital-intensive and more labour-intensive than general chemicals. Every campaign needs skilled operators, chemists and analysts.

Because so much cost is fixed, schedules matter. Some plants still run one or two shifts five days a week. A seven-day operation with four or five shift teams has become standard: extra pay for night work is more than offset by spreading fixed costs over more output.

Costs also fall with experience. Pollak's rule is that cost of goods should drop 10-20% each time yearly output doubles, through fewer steps, higher yields, cheaper raw materials, less solvent and recycled catalysts. Waste is part of the bill: Sheldon's E-factor, kilograms of waste per kilogram of product, is typically 5-50 for fine chemicals. GlaxoSmithKline set a 2015 target of 30 kg per kg of API, down from 100 kg in 2005.

• Pollak (2011) · pdf pp. 98-102, 91-92, 185
• Wittcoff (2004), Ch 1 · pdf pp. 46, 59

3.4 Conversion cost is priced per cubic metre-hour, so idle time is the enemy

A multipurpose plant makes many products with different throughputs, so its cost cannot be divided by 'tonnes per year'. Pollak's method has two parts.

First, a standard operating cost per cubic metre-hour. His example plant has $20 million of yearly operating cost, runs 7,920 hours (330 days of 24 hours), has 100 m3 of reactors and is 60% utilised. Available capacity is 7,920 x 100 x 0.6 = 475,200 m3-hours, so each costs $20,000,000 / 475,200 = $42.09.

Second, the volume-time-output (VTO) of a process step: reactor volume times hours, divided by good product. A step using three 3 m3 reactors (9 m3) for 30 hours to make 1,000 kg, with an operating factor of 0.9 for running-in, cleaning and learning, has VTO = 9 x 30 / (1,000 x 0.9) = 0.3 m3-hours per kg. Conversion cost is $42.09 x 0.3 = $12.63/kg, before raw materials. Pollak rates a VTO below 0.5 effective and above 1.0 ineffective. VTO also gives capacity: a three-step API with total VTO of 3 could be made at 475,200 / 3 = about 158,000 kg a year.

Utilisation sits in the denominator, so every idle hour raises the cost of every kilogram, as the table shows. It depends on a full order book and on using time well. Overall Equipment Effectiveness (OEE) multiplies availability, performance (actual versus theoretical speed) and quality (share of output released); Pollak's illustration is only about 33%. Time is lost to late materials, unstable processes, slow changeovers and waiting for tests; with short campaigns, as one Boehringer-Ingelheim manager put it, cleaning can become the plant's biggest 'product'.

• Pollak (2011) · pdf pp. 83-84, 98-100, 205-206, 271

See it in the table

Check your understanding

A bay of 9 m3 makes 500 kg of good product in a 30-hour batch with an operating factor of 0.9. What is its VTO, and how is it rated?

Why does Pollak's standard cost per m3-hour rise when utilisation falls?

Six 2 m3 reactors replace six 4 m3 reactors in Pollak's $21 million plant, with cost scaling with size to the power 0.5. What happens?