Rasipanje električnih parametara pojedinačnih primeraka
JFET-a u okviru jedinstvenog tipa zahteva ekstrahovanje
parametara SPICE modela ovih tranzistora na osnovu
eksperimentalnih strujno-naponskih karakteristika seta
uzoraka. U ovom radu su primenjene tri metodologije
određivanja parametara statičkog SPICE modela n-kanalnog
JFET-a J112. Eksperimentalna merenja odgovaraju uslovima
testiranja datim u tehničkim specifikacijama komponente
kako bi se izbeglo samozagrevanje. Verifikacija
pojedinačnih metodologija je sprovedena primenom modela u
LTspice-u i poređenjem dobijenih strujno-naponskih
karakteristika sa eksperimentalnim. Ukazano je na
najpogodniju metodologiju u zavisnosti od oblasti rada,
odnosno konkretnu primenu JFET-a.

У овом раду развијамо операторску формулацију
фотоакустичког одзива полупроводника, у којој се спрегнути
систем дифузионих и таласних једначина представља преко
резолвента у Лапласовом домену. Преносна функција која
повезује оптичку побуду и мерени акустички сигнал добија се
као пројекција резолвента на мерене величине, што омогућава
доследну анализу у временском домену. Добијени одзив има
структуру суперпозиције више изворних канала филтрираних
термалном и акустичком пропагацијом. Оваква формулација
омогућава раздвајање улоге пропагационих процеса и
механизама генерисања извора: доминантни полови су одређени
термалним и акустичким транспортом, док фотогенерисани
носиоци утичу на одзив кроз модификацију амплитуда и фаза
појединачних канала, што доводи до појаве нула у укупној
преносној функцији. Показујемо да облик преносне функције
зависи од карактеристичних просторних и временских скала
система, при чему се у одређеним лимитима може свести на
ефективне рационалне апроксимације. Овај приступ пружа
основу за систематску анализу различитих режима
(танак/дебео узорак, кратка/дуга времена) и указује да се
избором побуде — посебно поворком импулса са контролисаним
параметрима — може управљати релативним доприносима
појединачних канала и осетљивошћу на параметре носилаца.

This paper investigates a model of the electrical activity
of a single cell with two gating variables and three
critical thresholds (uv, uc, u_c^si), which introduce
additional discontinuous nonlinearities into the system
dynamics. It is shown that the resulting formulation
belongs to the class of coupled Filippov differential
inclusions and may exhibit non-uniqueness of solutions and
path-dependent transient behavior in the vicinity of the
thresholds. To provide a physical interpretation of these
phenomena, an equivalent electrical representation is
developed in terms of a network of capacitors, resistors,
and voltage-controlled current sources with feedback. The
analysis reveals that the model requires the introduction
of nonstandard voltage-controlled current sources involving
products of the membrane potential and gating variables, as
well as ideal switching elements. It is precisely the
combination of smooth nonlinearities and ideal thresholds
that leads to multivalued vector fields and non-unique
transient trajectories. To address this issue, the use of a
realistic diode current–voltage characteristic is proposed
in place of ideal switching elements. This modification
naturally introduces a finite transition time, which
physically corresponds to delays in the gating dynamics
and, mathematically, results in a smooth regularization of
the thresholds. As a consequence, the system transitions
from a differential inclusion framework to a system of
coupled nonlinear differential equations with a unique
solution. These results suggest that the observed
non-uniqueness in idealized models may stem from the
neglect of intrinsic time scales, and that their proper
incorporation leads to a more consistent and physically
realistic description of membrane potential dynamics.

This paper presents the fabrication and characterization of
RADFET radiation sensors with different gate dielectrics:
SiO2 (300 nm and 100 nm) and HfO2/SiO2 (40/5 nm). The
devices were irradiated using a 60Co gamma source up to a
total dose of 100 Gy. Electrical characterization was
performed through transfer characteristics,
transconductance, and threshold voltage extraction. The
results show a strong dependence of sensitivity on the gate
dielectric, with higher sensitivity observed for thicker
SiO2 layers compared to the HfO2-based structure.

This paper gives an overview of IHP activities within the
APECS project, a pilot line for heterogeneous integration
and advanced packaging technologies, focused on
strengthening Europe’s semiconductor manufacturing
capabilities and chiplet innovation. The activities include
the development of wafer-to-wafer and die-to-wafer
CMOS-to-BiCMOS Al-to-Al bonding technology, the development
of an open source design kit for the IHP Si-interposer
process technology, heterointegration of D-band and H-band
RF circuits, and the implementation of RISC-V based chiplet
for a multi-material sensor demonstrator.

This paper addresses the partial inverse problem in a
high-dimensional parametric forward two-layer system in
photoacoustics, aiming to estimate layer thickness and
optical absorption parameters from frequency-domain
amplitude and
phase responses. The proposed approach follows a
physics-based machine learning optimization procedure and
belongs to a class of model-based methods grounded in
analytical solutions. A neural network is trained to
predict the unknown parameters, while an analytical forward
model, derived from the governing equations, is embedded
into the loss function. The training objective combines a
data-driven component with a physics-based constraint that
enforces data consistency, an agreement between numerical
experiments and those reconstructed from the predicted
parameters through the forward operator. We evaluate the
inverse problem on a parametrized forward operator by
sampling its output over a defined parameter space. The
results demonstrate accurate and stable low-dimensional
parameter recovery under model overparameterization,
highlighting the effectiveness of integrating an analytical
model with data-driven optimization and proving inverse
optimization under model degeneracy.