The discriminating variable in the four-population bead experiment is one degree at each end of the channel. At 23-25 degrees C the 20.2 and 13.0 micron particles left through the same outlet while the 9.94 and 5.65 micron ones were collected separately; at 24-26 degrees C the two large populations split and the two small ones were pooled instead [9]. That is a 1 degree rise at both ends [12]. Note that the windows overlap: 24 and 25 degrees C appear in both [13]. So the outcome is not fixed by the temperature at any given pillar row. It is fixed by where along the array the swelling PNIPAM pillars narrow the gaps past a particle's critical diameter [6], and the endpoints of the gradient, imposed by two Peltier elements, decide where that crossing sits [7].
That is the real departure. In a conventional DLD array the critical diameter is cut into the pillar geometry, which is why isolating several fractions has meant several devices [4]. Here one array holds a range of thresholds at once, and the three-bead test came out at 92.9, 90.0 and 97.8 percent purity for the 13.0, 9.94 and 5.65 micron populations [8]. The tightest pair the array resolved differs by a factor of 1.31 [14], which is the number to hold onto, because overlapping size distributions are exactly what makes blood hard [11].
The blood run itself is where this account thins out. Cancer cells, white blood cells and red blood cells were isolated from diluted whole blood, and the cancer cells are described as viable [10]. No purity, recovery, throughput or viability percentage is given for that experiment in the phys.org write-up [17], and those are the figures that decide whether a rare-cell workflow can use this. A 1.31x resolution on polystyrene beads is not the same problem as separating leukocytes from a tumour cell of similar diameter but different stiffness.
What the design buys in flexibility, it spends on control. A geometry-defined threshold cannot drift; a thermally defined one can. Every reported condition lives inside a 3 degree C span, from 23 to 26 [15], and a single degree of error at one end is the same magnitude as the deliberate change that reprogrammed the device [12]. Gradient stability, Peltier calibration and the thermal load of the sample itself all become part of the separation spec rather than the fabrication spec. For a group that already runs one DLD chip per fraction, that trade is probably worth taking, since it replaces a set of masks with a controller. For anyone hoping to work near physiological temperature, the demonstrated window is a room-temperature one, and the source says nothing about behaviour above it.
The study, led by Takasi Nisisako with Yusuke Kanno and Ze Jiang, appeared online in Lab on a Chip on July 29, 2026 [2][3].
