The crystal structure of TiNiSn belongs to a cubic system, where each kind of atom forms a face centered cubic lattice, as shown in Fig. 1a. The XRD patterns of the synthesized TiZr*x*NiSn samples are presented in Fig. 1b, which shows the main diffraction peaks of all the samples could be indexed to the cubic phase TiNiSn (reference code: 03 − 065 − 0617), suggesting the samples could be successfully prepared by the modified solid-state reaction. Besides, a few weak peaks corresponding to TiNi2Sn, ZrNiSn, Sn5Ti6 and Sn were detected. The enlarged XRD patterns in the 2θ range of 42°−44° clearly show the low-angle shifting of major diffraction peak, (220) plane, with increasing Zr content, which is possibly due to the incorporation of the larger ions Zr4+ (inradius: 72 pm) into the Ti4+ (inradius༚53 pm) sublattice. As expect, Zr doping leads to a slight expansion of the unit cell, see the calculated lattice parameters in Table 1. And, the smaller electronegativity of Zr (1.33) compared with Ti (1.54) could lead to an increase in the bonding force, making the crystal structure more stable.

Figure 2 and S2 show the SEM images of freshly fractured surface for TiZr*x*NiSn (*x* = 0, 0.005, 0.015, 0.025) samples. It can be seen that each bulk sample is comprised of the compacted grains, which is consistent with its high relative density. Notably, the size of the grains in the pristine sample TiNiSn (*x* = 0) is on micrometer scale (Fig. 2a). Notably, as the Zr amount increases to *x* = 0.015, there appear many nanoparticles in the matrix (Fig. 2c), which is conducive to reducing the *κ**l* of the materials. Fig. S3 is the local particle size statistics of TiZr0.015NiSn. It can be seen that the particle diameter in this region ranges from 25nm to 726nm, of which 79.5% is less than 100 nm, and the average grain size is 83.7nm. To further clarify the composition of these nanoparticles, EDS mapping was carried out on the typical sample TiZr0.015NiSn, as shown in Fig. S4. It shows that the aggregation area of Zr and Sn is exactly corresponded to the nanoparticles accumulated zone, suggesting that these nanoparticles are in (Zr, Sn)-rich phase.

As shown in Fig. 3, TEM and HRTEM characterizations are performed on the pristine TiNiSn and TiZr0.015NiSn samples. Figure 3b shows a HRTEM image of the pristine TiNiSn sample taken from the marked area in Fig. 3a. It can be seen that the clear lattice fringes corresponding to plane spread over a large area (> 20 nm), showing a good crystallinity of the pristine sample. In contrast, as observed in Fig. 3d, there are nanodomains with a size of ~ 15 nm in the TiZr0.015NiSn matrix, which has a superlattice structure with the characteristic plane spacing of 8.66 Å, as it is coincident about 3 times of lattice spacings of plane , 2.94 Å. We confirmed that these nanodomains are uniquely dispersing in the current TiZr0.015NiSn matrix, as shown at black arrowed points in Fig. 4c by using HRTEM technique. Such nanodomains might also play an important role in suppressing the *κ**l*.

Figure 4a plots the temperature-dependent *n* of the samples TiZr*x*NiSn (*x* = 0, 0.005, 0.015, 0.025), which shows a decrease trend with the increase of Zr content. It is demonstrated that the impurity of residul elemental Sn could increase the *n* in TiNiSn [15, 31].Conversely, the residual Sn in TiNiSn is diminished with the increase of Zr content (Fig. 1b), which could contribute a decrease in the *n*.

Figure 4b polts the temperature-dependent *µ* of the samples TiZr*x*NiSn, which shows an increase trend with the increase of Zr content. This could be ascribed to to the significant decrease in *n* and the change of carrier scattering mechanism [32]. For the TiNiSn with nominal Zr content at *x* = 0 and 0.005, the carrier mobility is divided into two regions over the temperature range of 300 − 720 K. When the temperature is lower than 500 K, the mobility increases with rising the temperature, suggesting the scattering of ionized impurity is dominant. When the temperature above 500 K, the temperature-dependent *µ* obeys *T*− 0.5 law, which indicates that the alloy scattering is dominant, which is basically consistent with the literatures [33].

Figure 5 shows the temperature-dependent electrical properties of the samples TiZr*x*NiSn (*x* = 0, 0.005, 0.015, 0.025). Figure 5a demonstrates that the *σ* of the TiZr*x*NiSn samples increases with rising temperature, for the typical sample TiZr0.015NiSn, which is from 342 S cm− 1 at 323 K to 914 S cm− 1 at 823 K. And the electrical conductivity shows a decreasing tendency with increasing the Zr content, which decreases from 611 S cm− 1 for the pristine sample to 274 S cm− 1 for the sample TiZr0.025NiSn at room temperature, which is consistent with the change of carrier concentration. Additionally, it is confirmed that the Sn-rich metallic phase could result in the increase of *σ*, as shown in Fig. S5. Notably, the jump region in the temperature-dependent *σ* curve (marked in yellow) could be attributed to the elemental Sn resided in the matrix, which has a low melting point (505 K).

The temperature-dependent *S* of the TiZr*x*NiSn samples are plotted in Fig. 5b. The negative value indicates an *n*-type conducting behavior of TiZr*x*NiSn samples. From the Goldsmid formula [37], the band gap can be estimated by the maximum Seebeck coefficient and its corresponding temperature. It is found that Zr doping can improve the band gap from 0.19 eV to 0.25 eV. With rising temperature (> 526 K), the decrease of the absolute value of *S* for all samples could be attributed to the bipolar diffusion effect at high temperatures. Additionally, it can be seen that the absolute value of *S* increases with increasing the Zr content, which is mainly due to the decrease in *n*. For a degenerate semiconductor, the *S* is commonly correlated with *n*, which could be described by the Mott equation [38], *i*.*e*., , where *k**B* is the Boltzmann constant, *h* is the Planck constant, *m** is the carrier density of states effective mass, and *e* is the elemental charge.

Furthermore, the *S* as a function of *n* at 323 K, together with Pisarenko curve based on the single parabolic model using *m** = 1.52 *m**e*, is plotted in Fig. 5c. For the samples with *x* = 0 and 0.005, the absolute *S* agrees well with the fitted curve, while the absolute *S* for *x* = 0.015 and 0.025 samples locates below the curve, indicating the decrease in *m**, which is consistent with the increase of *µ* (Fig. 4b). With both decrease in *n* and *m**, it reaches a conclusion that the enhanced absolute *S* for Zr-doped samples is due to the decline in *n* according to the Mott equation. As expected, the maximum absolute value of *S* up to 207 µV K− 1 at 538 K is attained in the TiZr0.015NiSn sample.

Figure 5d shows the temperature-dependent *PF* of the samples TiZr*x*NiSn. The power factor is significantly improved in Zr doped samples due to the significant enhancement of *S*. The maximum power factor of 3.29 mW m− 1 K− 2 at 773 K is achieved in the sample TiZr0.015NiSn, which is 50% higher than that value of the pristine TiNiSn sample, and comparable to the Mn-doped TiNiSn [15].

Figure 6a shows the *κ* of TiZr*x*NiSn (*x* = 0, 0.005, 0.015, 0.025) samples. We can see that the pristine sample exhibits a high *κ*, 3.9 − 4.4 W m− 1 K− 1 in the temperature range of 323 − 823 K. And extra Zr doping could effectively reduce the *κ* of the TiNiSn-based material, as demonstrated in the doped samples. The lowest *κ* is obtained in the sample TiZr0.015NiSn, which is 2.3 W m− 1 K− 1 at 573 K. To clarify the reasons, we calculated the *κ**e* (*κ*e = *LσT*, where *L* is the Lorentz number determined by fitting the measured *S* based on the single parabolic band model [39, 40], as shown in Fig. S6c) and the *κ**l* (*κ**l* = *κ* − *κ**e*), which are plotted in Fig. S6d and Fig. 6b, respectively. It shows that the *κ* is mainly contributed by the *κ**l*, especially at low temperatures (< 600 K). Also the decrease in *κ* for the Zr-doped samples is mainly contributed by the reduction of *κ**l*. The minimum *κ**l* as low as 1.74 W m− 1 K− 1 at 723K is attained in the TiZr0.015NiSn. In order to further understand the contribution of various phonon scattering sources on the reduction of *κ**l*, the Debye-Callaway model is employed [41]. The solid line is the theoretical *κ**l* with respect to mass and strain fluctuation caused by Zr substitution on Ti sites. As shown in Fig. 6c, the experimental *κ**l* is significantly lower the theoretical curve, suggesting the reduction of *κ**l* is originated from other phonon scattering sources. According to the XRD, SEM, and TEM characterization results, it is believed that the secondary nanoparticles and superlattice-like nanodomains may be responsible for the low *κ**l* in TiZr0.015NiSn.

Figure 7a shows the temperature-dependent *ZT* of the samples TiZr*x*NiSn over 323 − 823 K, which are greatly improved by extra Zr doping. The maximum *ZT* value of 0.88 at 773 K is achieved in TiZr0.015NiSn, which is more than twice that value of the pristine sample. The average *ZT* value is calculated using the formula, where *T*c and *T*h are the temperatures of the cold and hot end, respectively, *ZT*avg is the ratio between integrated area under the *ZT* curve and temperature difference ΔT=Th-Tc. The average *ZT* value obtained in the sample TiZr0.015NiSn is up to 0.62 in the temperature range of 373−773 K, which is much higher than the reported values of the counterparts [14,15,24,42−44].