Crosspoint Switch
How a Crosspoint Matrix Routes Any Input to Any Output
The crosspoint name comes from the physical layout: input signals run along horizontal row buses, output ports tap vertical column buses, and a switch cell sits at every point where a row crosses a column. Closing a single cell connects that row to that column; leaving the rest open keeps every other path isolated. Because the cells are independent, the control logic can assign output 3 to input 7 without ever touching the connection feeding output 1. This is the operational meaning of a nonblocking matrix, and it is the property that distinguishes a true crosspoint from a tree or Clos network that shares internal nodes to save hardware.
Each cell is itself a small RF switch, implemented with PIN diodes, GaAs or silicon-on-insulator FETs, MEMS contacts, or electromechanical relays depending on speed, power, and frequency needs. Solid-state cells switch in nanoseconds and survive billions of cycles but add insertion loss and have finite off-state isolation; relay cells offer near-lossless paths and superb isolation but switch in milliseconds and wear out. The matrix designer trades these against the cell count, since an 8 by 8 full crosspoint already requires 64 cells and a 16 by 16 requires 256.
Broadcast is the second defining capability. Unlike a single-pole multiplexer that can route a source to only one destination, a crosspoint can close several cells on the same input row to fan one signal out to many outputs simultaneously, which is why broadcast video and audio routers and test-stimulus distribution networks rely on the topology.
Isolation and Crosstalk in Large Matrices
The dominant signal-integrity challenge is leakage onto shared buses. Every output column is touched by M off-state cells, and each leaks a small fraction of its input onto that column. The leaked contributions add, so a matrix built from cells with excellent single-cell isolation can still show mediocre matrix isolation once dozens of inputs share a bus. Series-shunt cell topologies, grounded guard traces between buses, and absorptive terminations on idle ports are the standard countermeasures, and very large routers are partitioned into buffered crosspoint tiles to rebuild isolation.
Governing Relationships
Ncells = Nin × Mout
Worst-case column isolation (M inputs sharing a bus):
ISOcol ≈ ISOcell − 20·log10(M − 1) dB
End-to-end insertion loss:
ILpath ≈ ILcell + Lrow bus + Lcol bus dB
Example: a 16-input column built from cells with ISOcell ≈ 60 dB gives ISOcol ≈ 60 − 20·log10(15) ≈ 36 dB before layout coupling. A solid-state cell with ILcell ≈ 1.5 dB plus 2 dB of bus loss yields ILpath ≈ 3.5 dB.
Crosspoint vs. Other Switch Architectures
| Architecture | Cells (8×8) | Nonblocking | Broadcast | Isolation Scaling | Best Application |
|---|---|---|---|---|---|
| Full crosspoint | 64 | Yes | Yes | Degrades with bus size | Broadcast routers, test distribution |
| Blocking tree (SPnT) | ~16 | No | No | Excellent (point to point) | Single-path selection |
| Clos network | ~40 | Rearrangeable | Limited | Good | Large telecom fabrics |
| Electromechanical matrix | 64 relays | Yes | Yes | Very high (>80 dB) | Low-loss instrumentation to 18 GHz |
Frequently Asked Questions
What is the difference between a crosspoint switch and a blocking switch matrix?
A full crosspoint places a dedicated cell at every row-column intersection, needing N×M cells, so any output can connect to any input without disturbing the others; that is the definition of nonblocking, and it also allows broadcast. A blocking matrix uses a tree or Clos arrangement of single-pole switches to save cells, but two desired connections can contend for the same internal node and force one to be rejected. Choose crosspoint when guaranteed nonblocking behavior and broadcast matter more than cell count.
How does isolation scale as a crosspoint switch matrix grows larger?
Matrix isolation degrades as the array grows because each output bus collects leakage from every off-state cell tied to it. With M inputs sharing a column, isolation drops by roughly 20·log10(M − 1) relative to one cell, so a 60 dB cell on a 16-input bus falls toward about 36 dB before layout coupling. Series-shunt cells, guard traces, and absorptive terminations on idle ports help, and large matrices are partitioned into buffered crosspoint tiles to restore isolation.
What insertion loss should I expect from a crosspoint switch path?
It depends on cell technology and how many series elements a signal crosses. Solid-state GaAs or silicon-on-insulator cells add roughly 0.5 to 2 dB per connection at 1 to 6 GHz, and since both the row bus and column bus add loss a large monolithic matrix can show 3 to 6 dB end to end at the band edge. Electromechanical relay matrices stay under about 0.3 dB to 18 GHz but switch in milliseconds rather than nanoseconds.