AI agent drives engineering design tools to create a 251 gigahertz wave transmitter
Analog circuit and radio-frequency engineering has remained a chronic obstacle to full automation, as designing a chip that operates at hundreds of gigahertz requires weeks of manual tuning and complex electromagnetic field simulation. A research paper accepted for presentation at the IEEE Radio and Wireless Week 2027 reveals a practical breakthrough of this barrier, where a new framework called "TACS-RF" succeeded in employing an AI-augmented software agent to directly drive standard electronic design automation tools, culminating in the generation of a fully integrated millimeter-wave transmitter operating at sub-terahertz frequencies.
The research framework, developed by researchers Tristan Liang and Aravind Nagulu, relied on the software model "Cloud Opus 4.8" to run industry-standard electronic design automation tool suites, rather than merely suggesting abstract code snippets. The agent guided, in an interactive and direct manner, the "Cadence Virtuoso" and "Spectre" tools for circuit simulation, the "Calibre" program for physical verification, as well as "Cadence EMX" and "Ansys HFSS" for extracting the electromagnetic parameters of passive components, forming a closed-loop feedback cycle that tests the design and automatically corrects its pathways.
The experiment demonstrated the efficiency of semi-autonomous design by fabricating and testing a transmitter operating in the 240-251 GHz band, using a 90-nm SiGe BiCMOS process. The agent designed the transmitter’s sub-blocks, including a voltage-controlled Colpitts oscillator at 60 GHz with second-order harmonic extraction, multiple differential power-switch stages, a frequency multiplier, a differential antenna with a neck-tie layout, and impedance-matching networks across all stages. After extracting parasitic effects and performing electromagnetic verification, the transmission chain achieved an output power ranging from 5.1 to 6.3 dBm over a tunable bandwidth of 11 GHz, phase noise of, 80 to, 88 dBc/Hz at a 1-MHz offset, and a continuous-wave power consumption of 361 mW.
This practical pathway represents a tangible step towardHardware automation from code to chipIn radio-frequency circuits, this defines a new division of labor between the human engineer and the intelligent agent. Instead of engineers spending time manually tweaking circuit elements and chasing recurring simulation errors, the human role is limited to defining high-level specifications and governing standards, while the agent carries out the design iterations, physical extraction, and iterative refinement that previously consumed months of specialized engineering effort.
This shift has direct implications for the advanced-technology ecosystem in the Gulf, Egypt, and the Levant, where regional semiconductor initiatives and tech hubs aim to establish chip-design companies without owning physical foundries. The scarcity of highly skilled RF engineers and the high cost of simulation tool licenses have long hindered competition in advanced communications and radar hardware, but moving manual tuning burdens to intelligent agents reduces prototype exploration costs and shortens the time to market for designs ready for tape-out. This reality demands a redefinition of the skill set required in regional labs, moving from traditional training on design-tool interfaces to mastering the construction of agent-based workflow environments and verifying the electromagnetic integrity of their outputs.
The real challenge remains transferring design agents from academic settings to commercial production lines, where high-frequency communications hardware tolerates no margin for error in silicon physical properties. The success of an intelligent agent in steering standard design tools to operate a transmitter at frequencies above 240 GHz demonstrates that analog hardware automation is no longer a distant theoretical horizon, but an engineering reality that reshapes what small development teams can achieve with limited human resources.