Darshan Jetly

What factors drive LO generation and distribution frequency choices, especially in systems with multiple TX/RX paths (communications or radar)? Let’s take a classic example: a 60–80 GHz transceiver with multiple TX and RX paths. A number of competing factors weigh into this decision, for example: -the PLL’s optimal operating frequency and achievable FoM -how much of the system must be modeled with high-accuracy EM extraction -transmission loss and signal quality versus frequency for on-chip interconnects -the cost, noise, and spur implications of frequency multiplication (if used) -and others that are often highly implementation-dependent While phase noise is often considered the most critical metric, area, power, and confidence in silicon-to-simulation correlation also strongly guide the choice of both PLL generation frequency and distribution frequency. For instance, choosing direct LO distribution at 80 GHz avoids the need for frequency multipliers in each TX/RX path, reducing local circuit complexity and area. Individual passive elements in the distribution path are physically smaller at these frequencies. However, distributing at such high frequencies comes with significantly higher transmission losses, and buffer design becomes much more constrained: it tends to push toward resonant or narrowband stages, and the distribution network itself turns into a sensitive RF block. By contrast, distributing the LO at a lower frequency (e.g. ~15 GHz) requires frequency multiplication somewhere in the system, increasing local power consumption and circuit complexity at the multiplication points. At the same time, the distribution becomes far more predictable: losses are lower, simpler buffer stages may be sufficient, routing is easier, and high-accuracy modeling over a large on-chip network is considerably more tractable. The two choices effectively trade off area, power, and complexity between the distribution network and the TX/RX blocks. Similar tradeoffs appear at the LO generation stage. Generating directly at 80 GHz means operating in a regime where passive Q — especially of capacitive elements — degrades with frequency, and modeling uncertainty can force designers to over-design bandwidth or power to ensure margin. Generating the LO at a lower frequency and relying on multiplication, particularly when combined with lower-frequency distribution, is therefore often the more robust choice in practice. Ultimately, deciding which tradeoffs weigh heavier when choosing the LO generation and distribution frequency — and the corresponding architecture — depends on many factors, and their relative importance changes from one design to another. Understanding these tradeoffs is critical to making an informed and effective choice.

11