In semiconductor work there is a clear dividing line between people who can say what is happening and people who can only say what happened.
The layer everyone treats as optional
Device physics has a reputation as academic background: something you cover once, then leave behind for tools and datasheets. The reputation survives because on a good day it is true. Circuits get designed, wafers get processed and products ship without anyone invoking carrier transport.
It stops being true the moment something does not work for a reason nobody expected.
A parameter drifts at one corner of the process. A device that passed qualification fails after eight months in the field. Yield drops on one tool and the wafer maps show a pattern nobody can read. A part behaves differently from the simulation and the simulation was run correctly.
Without the physics, every unexplained failure becomes trial and error, and in this industry the feedback loop on a trial is weeks.
That is the real argument for the training. Not elegance. Loop time. A team that can reason about mechanism eliminates hypotheses on a whiteboard in an afternoon. A team that cannot has to build each one.
What the direction covers
The scope: solid state physics, semiconductor materials, transistor operation, device characterisation and the physics of scaling.
Four areas.
Carriers and transport. How charge moves, what limits it, and why mobility, doping and temperature appear in every equation that matters.
Junctions and interfaces. Where devices are made and where most of them fail. The oxide interface, the metal contact, the heterojunction.
Device operation. Transistor families, what the terminals actually do, and where the textbook model stops describing the real device.
Characterisation and failure analysis. Electrical test, physical analysis, and the discipline of inferring an internal mechanism from external measurements.
Three things the datasheet does not tell you
Datasheets are honest documents that are read carelessly.
Typical is not guaranteed. A parameter given as typical is a central value from a sample of devices from a process at a point in time. It is not a limit and is not a specification. A design whose function depends on a typical value has no margin at all, and the failure arrives with a process change the manufacturer was under no obligation to announce.
The test conditions are load bearing. On state resistance quoted at twenty five degrees roughly doubles at operating temperature in many power devices. Capacitance quoted at a low bias falls sharply under working bias. Gain is quoted at one current and one temperature. Reading a number without reading the conditions line beneath it is the most common error in the field.
Absolute maximum ratings are not operating points. They are the boundary beyond which damage may occur. Designing to them is designing at the edge of the cliff and expecting the cliff to have a handrail.
None of this is obscure. All of it is routinely missed by engineers who were never given the physical picture that would make the numbers mean something.
Why reliability is physics on a slow clock
Semiconductor devices do not usually fail randomly. They fail through identified physical mechanisms that proceed slowly under stress and eventually cross a threshold.
The named mechanisms are well characterised: gradual degradation of the gate oxide under electric field, damage from energetic carriers near the drain, gradual migration of metal atoms in interconnect under high current density, threshold shift under bias and temperature, and mechanical fatigue in packaging as materials with different expansion coefficients are cycled.
Two consequences follow, and they are the reason this direction matters commercially.
Reliability can be modelled. These mechanisms have temperature and voltage dependencies. That is what makes accelerated life testing possible: stress the device harder than it will be stressed in service, observe the time to failure, and extrapolate using the known dependency. Done without understanding the mechanism, accelerated testing produces a number with no meaning, because the stress may have activated a mechanism that will never occur in service.
Reliability can be designed for. Derating voltage, limiting current density, controlling junction temperature and managing thermal cycling are all direct responses to specific mechanisms. Teams that know which mechanism dominates their application derate the right parameter. Teams that do not derate everything by a rule of thumb and pay for it in size and cost.
Where this sits in the domain
Semiconductor physics and devices is the second of nine directions in Astra Trainer's semiconductors, electronics and quantum domain. It sits underneath almost everything else in it: chip design and VLSI, semiconductor manufacturing and advanced packaging, photonics and power electronics all rest on the same device layer.
It also connects outward to advanced materials, where electronic and semiconductor materials is a direction in its own right, and to energy, where wide bandgap devices are changing power conversion. Partners often scope physics alongside manufacturing, because it is the fastest way to raise the diagnostic capability of an existing fab workforce. You can see the nine directions here.
Wide bandgap, and why it is not a drop-in
Silicon carbide and gallium nitride are used increasingly in power conversion, electric vehicles, industrial drives and radio frequency power, and are frequently introduced into designs as though they were silicon parts with better numbers.
They are not, and the physics explains why.
They switch much faster, which is the point and also the problem. Fast switching edges mean high rates of voltage change, which couple through parasitic capacitance, drive oscillation in layout inductance and create electromagnetic interference that silicon designs never produced. The circuit that worked with a silicon device frequently does not work with a faster one in the same footprint.
Gate drive requirements differ, and in some device families the acceptable gate voltage window is narrower than engineers used to silicon expect, with damage on one side and excess conduction loss on the other.
Thermal conductivity does not solve thermal design. A device that can operate hotter and sits in a smaller package concentrates the same heat into less area. The package and the board become the limit.
The failure and reliability picture differs. These are newer technologies in volume production and their degradation behaviour under long term field conditions is less established than silicon's. That is a reason for care in qualification, not a reason for avoidance, but it deserves stating plainly rather than being left out of the sales conversation.
The roles, named
Device engineers. Device design, modelling and characterisation.
Process integration engineers. Where physics meets manufacturing, and where most of the hard yield problems are solved.
Failure analysis engineers. Electrical and physical analysis, decapsulation, imaging, fault isolation.
Reliability engineers. Qualification, accelerated testing and lifetime modelling.
Characterisation and test engineers. Parametric test and the measurement methods behind every datasheet line.
Compact modelling engineers, building the device models the entire chip design industry simulates against.
Applications engineers for power and radio frequency devices, translating physics into circuit guidance.
Technology development engineers at manufacturers, working several process generations ahead.
Who can be trained into it
Process technicians and equipment engineers. The most direct conversion. They already work with these devices and these tools every day, and they observe effects constantly. What they are usually missing is the framework that connects a wafer map pattern to a physical cause.
Test and product engineers. Into characterisation and failure analysis. They hold the parametric data and often lack the model that explains its shape.
Physics and materials graduates outside the industry. Frequently the shortest path, since the foundation is present and only the industrial context is missing.
Electronics engineers. Into device applications work, particularly for power and wide bandgap devices, where circuit knowledge plus device physics is a scarce combination.
Analytical laboratory staff. Into physical failure analysis, where microscopy and sample preparation skill transfers almost intact.
Quality engineers. Into reliability, where statistical method is already in place and the missing piece is mechanism.
Laboratory and fab hazards. Semiconductor processing and device analysis involve highly toxic and pyrophoric gases, strong acids including hydrofluoric acid, high voltage test equipment and ionising radiation sources. These carry serious and in some cases immediately life threatening risk, and are governed by specific regulatory and site requirements, controlled procedures and supervised training. Astra Trainer builds technical understanding and analytical capability. It does not provide the hands-on hazard training, authorisation or site qualification that working in these environments requires.
What to take from this
Device physics is not background. It is the layer that turns an unexplained failure into a short list of testable hypotheses.
Datasheet numbers mean what their conditions say they mean, and typical values are not specifications.
Reliability failures follow known mechanisms with known dependencies, which is exactly why they can be accelerated, modelled and designed against.
Wide bandgap devices are a different design problem, not a faster silicon part, and their long term field behaviour is less established.
And the people who see these effects most often already work in the fab. They are usually one conceptual layer from being able to explain them.
Why does device physics matter if the tools handle it?
Because tools describe expected behaviour. When something unexpected happens, physics is what narrows the possibilities without building each hypothesis, and in semiconductors a built hypothesis takes weeks.
What is the most common datasheet mistake?
Reading a parameter without its test conditions. On state resistance, capacitance and gain all change substantially with temperature and bias, and typical values are sample statistics rather than guarantees.
Why can semiconductor reliability be accelerated?
Because the dominant failure mechanisms are slow physical processes with known temperature and voltage dependencies. Stress the device harder, measure time to failure, extrapolate along the known dependency. Without knowing the mechanism, the extrapolation is meaningless.
Are silicon carbide and gallium nitride simply better than silicon?
They are better on several parameters and different on many others. Faster switching creates interference and oscillation problems, gate drive windows can be narrower, smaller packages concentrate heat, and long term field reliability data is less mature.
Who converts well into device physics roles?
Process technicians and equipment engineers first, since they observe these effects daily. Then test and product engineers into characterisation, physics graduates into device work, and analytical laboratory staff into physical failure analysis.