Impurities in Silicon Ingots: Oxygen, Carbon and Metals

Polysilicon purity is only the starting point

A silicon ingot picks up impurities after the feedstock is charged. The melt sits for many hours in a quartz crucible, inside a graphite hot zone, under flowing argon. Each of those materials can add something to the crystal, so a clean polysilicon certificate does not settle the question of ingot purity.

Each impurity also behaves differently once in the melt. Some are lifted away as gas and some build up as the melt is consumed. The sections below take the three groups in turn.

Oxygen: from the crucible

During growth, oxygen dissolves from the crucible into the melt and forms silicon monoxide (SiO), which the argon flow has to carry away, according to Lan and co-authors in Crystal Growth of Silicon for Solar Cells. Some of the oxygen stays in the crystal.

Its effect depends on the cell. In boron-doped p-type Cz wafers, interstitial oxygen reacts with boron to form the boron-oxygen defect, which PV-Manufacturing.org describes as the primary source of carrier-induced degradation in that material. In n-type silicon, the same book notes that oxygen precipitates usually reduce minority-carrier lifetime. Oxygen is not purely harmful, however: PV-Manufacturing.org notes that a relatively high oxygen level may assist internal gettering of impurities. Our guides on oxygen precipitation and magnetic Czochralski growth go deeper.

Carbon: mostly from the graphite hot zone

The SiO leaving the melt does not simply disappear. Researchers writing for the Electrochemical Society describe carbon contamination during Cz growth as mainly influenced by back-diffusion of carbon monoxide (CO) into the melt, with the CO generated when SiO reacts with graphite parts. The same paper reports that the starting polysilicon also contains carbon, at about 5 × 10^14 atoms/cm³, so feedstock and furnace both count.

Carbon matters mainly through oxygen. That paper reports that carbon acts as a heterogeneous nucleation site for oxygen precipitates and so shortens carrier lifetime. It was written about high-power semiconductor devices rather than solar cells, so the size of the effect in a solar wafer should be tested rather than assumed.

• Reduce the SiO and graphite reaction through hot-zone design. Nagai and co-authors found this effectively lowered the CO contamination rate.
• Raise the argon flow velocity above the melt, which the same authors found increased CO evaporation during melting and growth.
• Use FT-IR for routine checks, but note its reported carbon detection limit of about 10^15 atoms/cm³. Photoluminescence-based methods can reach the order of 10^13 atoms/cm³.

Metals: small amounts, large effects

Transition metals can do damage at very low concentrations. A study led by Coletti added iron, chromium, nickel, titanium and copper to feedstock on purpose, using p-type multicrystalline ingots. Iron, chromium and titanium reduced the diffusion length. Nickel did not reduce it significantly, but strongly affected emitter recombination. Copper affected both base-bulk and emitter recombination. Separate work presented at the 2007 European PV conference saw diffusion length fall in monocrystalline cells at the lowest iron level tested, 10^12 atoms/cm³.

Better cells have less tolerance. Coletti's later modelling ranked titanium as the highest-impact impurity, followed by copper, chromium, nickel and iron, and concluded that advanced devices will generally be more sensitive to impurity content. Thinner wafers partly offset this. That is a useful warning for TOPCon and heterojunction buyers, although the original data came from a different cell process, so limits should not be copied across cell types.

Segregation decides where metals end up. An impurity with a low segregation coefficient mostly stays in the melt, so it concentrates towards the end of the ingot. Recharging adds another concern: Lan and co-authors point to metal contamination during recharge and impurity accumulation in the crucible as the problems of multiple charges for semiconductor applications. PV-Manufacturing.org's ingot impurity calculator lets engineers model the axial profile.

Where HPQ sand and crucible quality fit

Because the crucible wall slowly dissolves during the run, whatever is in its inner layer has a path into the melt. That is why crucible buyers look at the purity of the high-purity quartz sand behind the inner layer as well as the crucible's dimensions. How much of a given impurity reaches the finished crystal depends on the element and the process, so the useful test is a qualification pull on the buyer's own puller, compared against a baseline.

What cell makers and ingot buyers can check

We did not find a single public table that sets acceptable limits for every cell. A practical approach is to write the tests into the specification, then agree limits with the supplier for the specific cell process.

• Resistivity along the ingot, which tracks dopant segregation.
• Minority-carrier lifetime on the ingot or wafer, and photoluminescence imaging to locate low-lifetime regions.
• FT-IR for oxygen and carbon, with a more sensitive method if carbon limits sit near the FT-IR detection limit.
• Trace-metal analysis of feedstock and sand, with the method and detection limit written down.
• Records of crucible lot, hot-zone age and recharge count, so that a lifetime drop can be traced to its cause.